Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Someone vibe coded a 3D pixel art city from real GitHub data. That someone was me. 29 days later: - 66,000+ developers in the city - 120K unique visitors - 449K pageviews - 2.2M ad impressions - 40+ brands advertising inside it - 4,238 GitHub stars - 3M+ organic... show more

Git City

1,101 subscribers

24,858 views • 5 months ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

A GUY AT GOOGLE DEEPMIND MADE AN ISOMETRIC PIXEL-ART MAP OF NEW YORK CITY AND PUT IT ON THE OPEN WEB FOR FREE it's called you open the tab and the city is just sitting there in classic SimCity 2000 isometric pixel art. you scroll. and it keeps going. and going. i zoomed in on midtown and i could read the H&M signage in times square. in red. as actual pixel-art letters on the side of a building. i could see the crystalline spire of the Bank of America Tower poking out of a clump of skyscrapers. individual rooftop HVAC units. tiny green roof gardens. the little driveway loops in front of the hotels. he estimates the map needs roughly 40,000 tiles. nothing is a placeholder. the guy who made it is Andy Coenen, a senior staff engineer at Google DeepMind. he is not a pixel artist. by his own admission he is "a former electronic musician." what he actually did is kind of insane: > pulled NYC's geometry from the Google Maps 3D tiles API > fine-tuned an open-source image model (Qwen-Image-Edit) on ~40 hand-paired examples of "satellite tile → pixel art tile" > spun up 50 parallel instances on rented GPUs and generated tens of thousands of tiles in a few hours > the fine-tune cost him 12 bucks his own stated mission for the project, verbatim, is one sentence: "what's possible now that was impossible before?" apparently the answer is "one engineer can pixel-art most of a metropolis for the price of a sandwich." and the wildest part to me is he didn't sell it. no signup. no paywall. no NFT. you open the URL and the city is yours to wander. the post landed at 1,325 points on Hacker News and topped bestofshowhn's 2026 list. we live in a timeline where a senior engineer at one of the largest AI labs on earth spent his nights pixel-arting Manhattan for fun and then gave it away. the internet is healing.

A GUY AT GOOGLE DEEPMIND MADE AN ISOMETRIC PIXEL-ART MAP OF NEW YORK CITY AND PUT IT ON THE OPEN WEB FOR FREE it's called you open the tab and the city is just sitting there in classic SimCity 2000 isometric pixel art. you scroll. and it keeps going. and going. i zoomed in on midtown and i could read the H&M signage in times square. in red. as actual pixel-art letters on the side of a building. i could see the crystalline spire of the Bank of America Tower poking out of a clump of skyscrapers. individual rooftop HVAC units. tiny green roof gardens. the little driveway loops in front of the hotels. he estimates the map needs roughly 40,000 tiles. nothing is a placeholder. the guy who made it is Andy Coenen, a senior staff engineer at Google DeepMind. he is not a pixel artist. by his own admission he is "a former electronic musician." what he actually did is kind of insane: > pulled NYC's geometry from the Google Maps 3D tiles API > fine-tuned an open-source image model (Qwen-Image-Edit) on ~40 hand-paired examples of "satellite tile → pixel art tile" > spun up 50 parallel instances on rented GPUs and generated tens of thousands of tiles in a few hours > the fine-tune cost him 12 bucks his own stated mission for the project, verbatim, is one sentence: "what's possible now that was impossible before?" apparently the answer is "one engineer can pixel-art most of a metropolis for the price of a sandwich." and the wildest part to me is he didn't sell it. no signup. no paywall. no NFT. you open the URL and the city is yours to wander. the post landed at 1,325 points on Hacker News and topped bestofshowhn's 2026 list. we live in a timeline where a senior engineer at one of the largest AI labs on earth spent his nights pixel-arting Manhattan for fun and then gave it away. the internet is healing.

Nav Toor

441,151 views • 2 months ago

Someone vibe coded a basketball hoop that floats above your Mac screen so you can shoot hoops while your agent codes. → The entire app is 1MB with no permissions or network access needed → Two modes, a 24-second shot clock where buzzer beaters count and free play → Your live score and all-time best sit at the top of the screen next to the notch → Floats above everything on the desktop so you never leave what you're doing → Free and open source The net was the hardest part. Claude rewrote it as a full cloth simulation. Then someone forked it and shipped a native Windows version at 469KB with real gravity constants, quadratic air drag, ball mass from the official rulebook and floor bounce from FIBA rebound data on a fixed 240Hz integrator. The golf version is already in progress. The notch is the hole.

Someone vibe coded a basketball hoop that floats above your Mac screen so you can shoot hoops while your agent codes. → The entire app is 1MB with no permissions or network access needed → Two modes, a 24-second shot clock where buzzer beaters count and free play → Your live score and all-time best sit at the top of the screen next to the notch → Floats above everything on the desktop so you never leave what you're doing → Free and open source The net was the hardest part. Claude rewrote it as a full cloth simulation. Then someone forked it and shipped a native Windows version at 469KB with real gravity constants, quadratic air drag, ball mass from the official rulebook and floor bounce from FIBA rebound data on a fixed 240Hz integrator. The golf version is already in progress. The notch is the hole.

0xMarioNawfal

40,904 views • 1 month ago

Anthropic's most viral feature is now open-source! Until now, Anthropic's Generative UI capabilities only existed inside its own products. CopilotKit🪁 just shipped Open Generative UI, an open-source implementation of Claude Artifacts that works in any app. The agent generates HTML/SVG at runtime, and CopilotKit streams it token-by-token into a sandboxed iframe inside the app's chat. So the user can watch the UI assemble itself in real time, not after the full response is ready. The sandbox is fully isolated with no access to the parent app, the DOM, or user data. So if the agent hallucinates broken markup or unexpected JavaScript, nothing leaks outside the iframe. Under the hood, the agent does not select from pre-built components. Instead, it generates arbitrary visuals from scratch every time. The output is unconstrained by default, but you can shape it by defining prompt-based skills that teach the agent specific visual formats or guidelines. For instance, a skill prompt can guide the agent toward producing a Chart.js dashboard with proper axis labels and responsive sizing, or an interactive 3D model with rotation controls. The video below shows this in action, and the output quality you see actually comes from the skills layer. Open Generative UI runs on AG-UI, so it works out of the box with LangGraph, CrewAI, Mastra, Google ADK, AWS Strands, and more. It also ships with a standalone MCP server that plugs into Claude Code, Cursor, or any MCP-compatible client. And the entire stack is built on top of CopilotKit, the open-source frontend framework for agents and generative UI. 30k+ GitHub stars, with SDKs for React, Next.js, Angular, and Vue. I have shared the GitHub repo and a live playground in the replies!

Anthropic's most viral feature is now open-source! Until now, Anthropic's Generative UI capabilities only existed inside its own products. CopilotKit🪁 just shipped Open Generative UI, an open-source implementation of Claude Artifacts that works in any app. The agent generates HTML/SVG at runtime, and CopilotKit streams it token-by-token into a sandboxed iframe inside the app's chat. So the user can watch the UI assemble itself in real time, not after the full response is ready. The sandbox is fully isolated with no access to the parent app, the DOM, or user data. So if the agent hallucinates broken markup or unexpected JavaScript, nothing leaks outside the iframe. Under the hood, the agent does not select from pre-built components. Instead, it generates arbitrary visuals from scratch every time. The output is unconstrained by default, but you can shape it by defining prompt-based skills that teach the agent specific visual formats or guidelines. For instance, a skill prompt can guide the agent toward producing a Chart.js dashboard with proper axis labels and responsive sizing, or an interactive 3D model with rotation controls. The video below shows this in action, and the output quality you see actually comes from the skills layer. Open Generative UI runs on AG-UI, so it works out of the box with LangGraph, CrewAI, Mastra, Google ADK, AWS Strands, and more. It also ships with a standalone MCP server that plugs into Claude Code, Cursor, or any MCP-compatible client. And the entire stack is built on top of CopilotKit, the open-source frontend framework for agents and generative UI. 30k+ GitHub stars, with SDKs for React, Next.js, Angular, and Vue. I have shared the GitHub repo and a live playground in the replies!

Akshay 🚀

87,048 views • 4 months ago

an ex-OpenAI researcher pushed his trading system to GitHub the night he got fired someone in a crypto Discord dropped the link at 2am. "this repo appeared an hour ago. zero stars. readme says internal use only. the code looks real" i clicked. it was a complete prediction market execution engine. bayesian consensus scoring. whale divergence tracker. kelly position sizing. edge detection on 5-minute BTC binary markets. clean Python. commented like someone who writes for a team. i ran git clone before finishing the readme. by 4am the repo was gone. deleted. commits. branches. history. everything. but i had it. opened a Chinese AI. gave it the code and one prompt. "adapt this for Polymarket. deploy it live. keep the scoring logic intact" 18,000 stars. open source HFT framework. the AI used it as the execution backbone. order routing. position management. risk controls. merged the stolen scoring model with hummingbot's engine. system was live by sunrise. 288 BTC markets per day. every 5 minutes. the crowd bets direction. his model doesn't care about direction. it buys whichever side the crowd underpriced. pair cost below 99c = profit regardless of outcome. the part that scared me: it worked immediately. first trade. green. second trade. green. third. fourth. fifth. all green. like it was built by someone who actually knew what they were doing. because it was. edge table live: BTC+100K 15m +7.5c. BTC+103K 30m spread 7c. BTC+101K +12.0c. BTC 0xd8dA...6045 +$9,260. 92%. > 0xBE0e...48CD +$7,147. 85%. > 0x47ac...d563 +$5,874. 81%. 323 trades. 85% win rate. sharpe 3.96. drawdown -0.9%. +$19,812 from $1,100 seed. my setup: > AI - $20/mo > VPS - $5/mo > hummingbot - free > his code - free now copy setup you can here: checked his GitHub profile last week. still active. new repos. normal stuff. open source contributions. no mention of the deleted repo anywhere. he has no idea someone cloned it four minutes before it disappeared. and now you're reading what was in it.

Hanako

29,053 views • 4 months ago

UPDATE: Ended up with this cool little tool to both build and verify the training data quality; it turned out very accurate! So I am at least happy about that (tho it obvisouly was the easy part) The pipeline to produce this kind of training data (from any Warsh recitation) is in the GitHub below. The repo also documents my full exploration of the Tarteel-for-Warsh problem (and quite frankly, of the offline open source tarteel problem in general): what I tried, what's promising, what's not, and what I think the next steps worth taking are: from a non-engineer with little technical expertiste trying to figure it out. I tried to document everything so the repo might be excessively comprehensive at times. It's basically my entire convo with Claude on the subject, condensed in one repo. Please someone take this and run with it! I genuinely believe the right engineers who can make something real out of this are out there First public repo I've ever shared, so if I made dumb mistakes, please tell me! 🔗

UPDATE: Ended up with this cool little tool to both build and verify the training data quality; it turned out very accurate! So I am at least happy about that (tho it obvisouly was the easy part) The pipeline to produce this kind of training data (from any Warsh recitation) is in the GitHub below. The repo also documents my full exploration of the Tarteel-for-Warsh problem (and quite frankly, of the offline open source tarteel problem in general): what I tried, what's promising, what's not, and what I think the next steps worth taking are: from a non-engineer with little technical expertiste trying to figure it out. I tried to document everything so the repo might be excessively comprehensive at times. It's basically my entire convo with Claude on the subject, condensed in one repo. Please someone take this and run with it! I genuinely believe the right engineers who can make something real out of this are out there First public repo I've ever shared, so if I made dumb mistakes, please tell me! 🔗

Yousr

38,068 views • 6 months ago

Anthropic's in trouble, again! They spent years building what's now fully open-source. What made Claude feel different from a normal app is that the agent could act inside the interface instead of only talking in a chat box. For instance, Claude Artifacts let an agent render real UI, charts, dashboards, and interactive components that assemble live inside the response. Every major AI product tried to replicate it. But the problem was that unlike reasoning, planning, tool-calling, etc., none of it shipped natively with LangGraph, CrewAI, or Google ADK. So teams started building an owned version that required engineering the entire interface layer from scratch. Most teams, however, just settled for shipping the agent as a backend API in a chat box since rendering the UI is only one piece of it. To actually make it work, the interface layer also needed real-time streaming, state kept in sync between agent and UI, conversations that persist across sessions, and reconnection when a user refreshes mid-run. CopilotKit🪁 is now the only open-source framework that actually lets you build your own full-stack Claude-like apps. It decouples the agent from the interface, talking over AG-UI (an open protocol for agent-to-user communication). Being a standard protocol, the frontend never needs to know whether it is talking to a LangGraph or a CrewAI agent. You can change the backend anytime and the UI will never notice. In practice, CopilotKit's interface layer gives several pre-implemented React building blocks that wire the agent directly into the app, like: - generative UI, so the agent renders real components instead of text - chat windows, sidebars, and popups, or a fully headless setup - shared state, so the agent and app stay in sync - human-in-the-loop approvals, where the agent waits before acting - persistent threads that store the whole session, including the agent-user interactions and generated UI, not just text And because that full history is captured, those interactions can feed a self-learning layer that also improves the agent from real usage over time. The interface layer that Anthropic spent years engineering in-house is now literally available to any developer/team. CopilotKit is open-source with 30k+ GitHub stars, and AG-UI, the protocol underneath, is already supported across every major agent framework: LangGraph, CrewAI, Mastra, Google ADK, and more. CopilotKit GitHub repo → (don't forget to star it ⭐ ) If you want to go deeper, I found a detailed breakdown by Shubham Saboo recently on the three Generative UI patterns, with implementation. Read it below.

Anthropic's in trouble, again! They spent years building what's now fully open-source. What made Claude feel different from a normal app is that the agent could act inside the interface instead of only talking in a chat box. For instance, Claude Artifacts let an agent render real UI, charts, dashboards, and interactive components that assemble live inside the response. Every major AI product tried to replicate it. But the problem was that unlike reasoning, planning, tool-calling, etc., none of it shipped natively with LangGraph, CrewAI, or Google ADK. So teams started building an owned version that required engineering the entire interface layer from scratch. Most teams, however, just settled for shipping the agent as a backend API in a chat box since rendering the UI is only one piece of it. To actually make it work, the interface layer also needed real-time streaming, state kept in sync between agent and UI, conversations that persist across sessions, and reconnection when a user refreshes mid-run. CopilotKit🪁 is now the only open-source framework that actually lets you build your own full-stack Claude-like apps. It decouples the agent from the interface, talking over AG-UI (an open protocol for agent-to-user communication). Being a standard protocol, the frontend never needs to know whether it is talking to a LangGraph or a CrewAI agent. You can change the backend anytime and the UI will never notice. In practice, CopilotKit's interface layer gives several pre-implemented React building blocks that wire the agent directly into the app, like: - generative UI, so the agent renders real components instead of text - chat windows, sidebars, and popups, or a fully headless setup - shared state, so the agent and app stay in sync - human-in-the-loop approvals, where the agent waits before acting - persistent threads that store the whole session, including the agent-user interactions and generated UI, not just text And because that full history is captured, those interactions can feed a self-learning layer that also improves the agent from real usage over time. The interface layer that Anthropic spent years engineering in-house is now literally available to any developer/team. CopilotKit is open-source with 30k+ GitHub stars, and AG-UI, the protocol underneath, is already supported across every major agent framework: LangGraph, CrewAI, Mastra, Google ADK, and more. CopilotKit GitHub repo → (don't forget to star it ⭐ ) If you want to go deeper, I found a detailed breakdown by Shubham Saboo recently on the three Generative UI patterns, with implementation. Read it below.

Avi Chawla

460,282 views • 3 months ago

Adobe tried to buy Figma for $20 billion in 2022. The deal collapsed. So Figma went public on the NYSE in July 2025 instead. Ticker FIG. Public company. Quarterly earnings. Wall Street pressure. You know what happens to design tools after they IPO. In March 2025, Figma raised the Professional Full seat 33%. From $15 to $20 a month. Organization seats jumped to $55. Enterprise to $90. Then they took Dev Mode, which was free during beta, and locked it behind a paid seat. Your developers now pay extra to inspect the designs your designers already paid to create. In March 2026, Figma started charging for AI credits on top. If Figma raises prices again, you pay. If Figma gets acquired, you pray. If Figma shuts down, your files die with it. Your design system. On their servers. In a proprietary format only their app can read. To draw rectangles on a screen. There is an open source design platform that runs on your hardware. Stores your files in plain SVG. Costs $0 forever for unlimited users. It is called Penpot. 45,700+ stars on GitHub. A full Figma-grade design platform built on open web standards. Vector editing. Components. Design tokens to W3C spec. Flex and Grid layouts. Real-time multiplayer. Interactive prototyping. Here's what it does: → Real-time collaboration. Live cursors. Comments in line. → Components, variants, shared libraries. → Auto layout, Flex, CSS Grid. The tool outputs production CSS, not lookalike CSS. → Interactive prototypes with overlays, animations, and flows. → Inspect tab. Free. Built in. Every developer grabs production CSS, SVG, HTML without a separate seat. → Plugin ecosystem. Figma import to migrate your files. → Self-host on Docker in one command. Your designs never leave your network. Here's the wildest part: Figma stores your designs in a proprietary format only Figma can read. Penpot files are SVG. The same format your browser has rendered for 25 years. Open them in any editor. Open them in 20 years. Nobody can lock you out. The feature Figma charges your developers extra for, Penpot gives away. Without asking permission. Figma Professional: $20/month per seat. A 10-person team: $2,400/year. Figma Organization: $55/month per Full seat. A 50-person org: $33,000/year. Penpot: $0. Unlimited users. Unlimited files. Unlimited teams. Self-hosted. Free forever. 45,700+ stars. 2,700+ forks. 250+ contributors. MPL-2.0 license. Backed by a community that believes design tools should be free. Your designs. Your files. Your standards. 100% Open Source. (Link in the comments)

Adobe tried to buy Figma for $20 billion in 2022. The deal collapsed. So Figma went public on the NYSE in July 2025 instead. Ticker FIG. Public company. Quarterly earnings. Wall Street pressure. You know what happens to design tools after they IPO. In March 2025, Figma raised the Professional Full seat 33%. From $15 to $20 a month. Organization seats jumped to $55. Enterprise to $90. Then they took Dev Mode, which was free during beta, and locked it behind a paid seat. Your developers now pay extra to inspect the designs your designers already paid to create. In March 2026, Figma started charging for AI credits on top. If Figma raises prices again, you pay. If Figma gets acquired, you pray. If Figma shuts down, your files die with it. Your design system. On their servers. In a proprietary format only their app can read. To draw rectangles on a screen. There is an open source design platform that runs on your hardware. Stores your files in plain SVG. Costs $0 forever for unlimited users. It is called Penpot. 45,700+ stars on GitHub. A full Figma-grade design platform built on open web standards. Vector editing. Components. Design tokens to W3C spec. Flex and Grid layouts. Real-time multiplayer. Interactive prototyping. Here's what it does: → Real-time collaboration. Live cursors. Comments in line. → Components, variants, shared libraries. → Auto layout, Flex, CSS Grid. The tool outputs production CSS, not lookalike CSS. → Interactive prototypes with overlays, animations, and flows. → Inspect tab. Free. Built in. Every developer grabs production CSS, SVG, HTML without a separate seat. → Plugin ecosystem. Figma import to migrate your files. → Self-host on Docker in one command. Your designs never leave your network. Here's the wildest part: Figma stores your designs in a proprietary format only Figma can read. Penpot files are SVG. The same format your browser has rendered for 25 years. Open them in any editor. Open them in 20 years. Nobody can lock you out. The feature Figma charges your developers extra for, Penpot gives away. Without asking permission. Figma Professional: $20/month per seat. A 10-person team: $2,400/year. Figma Organization: $55/month per Full seat. A 50-person org: $33,000/year. Penpot: $0. Unlimited users. Unlimited files. Unlimited teams. Self-hosted. Free forever. 45,700+ stars. 2,700+ forks. 250+ contributors. MPL-2.0 license. Backed by a community that believes design tools should be free. Your designs. Your files. Your standards. 100% Open Source. (Link in the comments)

Nav Toor

216,943 views • 4 months ago

Karpathy's prediction about RL is coming true now! He called reward functions unreliable and argued that a single reward number is too low-dimensional to teach an agent what "good" means for complex tasks. To solve this, Agents need a knowledge-guided review as a higher-dimensional feedback channel. Every major AI lab trains models with RL today (OpenAI, Anthropic, DeepSeek). And their key bottleneck has always been the reward functions. GRPO by DeepSeek worked well for math and code because the environment gave a binary signal. But for real agent tasks, someone still has to hand-code the scoring function. That takes days and breaks every time the pipeline changes. RULER (implemented in OpenPipe ART, 10k stars) addresses the exact problem Karpathy identified. The reward criteria are defined in plain English, and an LLM evaluates each trajectory against that description to provide feedback for training. I trained a Qwen3 1.4B agent that plays 2048 using GRPO with this exact workflow. In this case, the agent saw the board, picked a direction, and RULER evaluated the outcome, all from this natural language definition. You can see the full implementation on GitHub and try it yourself. Here's the ART Repo: (don't forget to star it ⭐ ) Just like RLHF replaced manual rankings and GRPO replaced the critic model, natural language rewards are replacing hand-coded scoring functions. RL reward engineering is now prompt engineering. I wrote a full walkthrough covering RL for LLM agents, from RLHF to GRPO to RULER, in the article below.

Karpathy's prediction about RL is coming true now! He called reward functions unreliable and argued that a single reward number is too low-dimensional to teach an agent what "good" means for complex tasks. To solve this, Agents need a knowledge-guided review as a higher-dimensional feedback channel. Every major AI lab trains models with RL today (OpenAI, Anthropic, DeepSeek). And their key bottleneck has always been the reward functions. GRPO by DeepSeek worked well for math and code because the environment gave a binary signal. But for real agent tasks, someone still has to hand-code the scoring function. That takes days and breaks every time the pipeline changes. RULER (implemented in OpenPipe ART, 10k stars) addresses the exact problem Karpathy identified. The reward criteria are defined in plain English, and an LLM evaluates each trajectory against that description to provide feedback for training. I trained a Qwen3 1.4B agent that plays 2048 using GRPO with this exact workflow. In this case, the agent saw the board, picked a direction, and RULER evaluated the outcome, all from this natural language definition. You can see the full implementation on GitHub and try it yourself. Here's the ART Repo: (don't forget to star it ⭐ ) Just like RLHF replaced manual rankings and GRPO replaced the critic model, natural language rewards are replacing hand-coded scoring functions. RL reward engineering is now prompt engineering. I wrote a full walkthrough covering RL for LLM agents, from RLHF to GRPO to RULER, in the article below.

Avi Chawla

350,512 views • 3 months ago

JustHireMe V0.1.31 is out. 🔗 Github: 📥 Download: The first post hit 390K impressions, 3.2K likes, 6K+ bookmarks, 1.2k+ GitHub stars and 400+ downloads . Genuinely didn't expect that. I've been shipping updates since and this release is a big one. Full video walkthrough attached but here's what's new: Command Center — proper dashboard now — active leads, scores, agent status, top roles — all in one screen — no more clicking around to figure out where things stand Customize One Job — paste any job URL — agent captures the JD, scores your fit, shows keyword gaps — generates a tailored resume + cover letter as PDFs — one page, one flow Outreach Generator — every lead now comes with ready-to-copy messages — 3-line founder DM — LinkedIn connection note — cold email — all written from your real profile, not templates Knowledge Graph (rebuilt from scratch) — Relation Atlas - weighted ribbons showing how projects prove skills — Gravity Graph - force-directed career layout — 3D Orbit - rotate through your profile in 3D — Curated mode - strongest evidence paths only — this is what the matching engine actually uses to rank you Job Pipeline — stages: Hot → New → Rated → Ready → Applied → Discarded — Fit score + Quality score on every lead — search, filter, re-eval, bulk clean, export — feels like a real job search CRM now Live Activity Stream — "what is the agent thinking?" — real-time logs of every query, eval, and decision — full transparency, nothing hidden Built-in AI Assistant — ask the app how to configure sources, understand a lead, generate packages — right inside the dashboard Modular Agent Config — Scout, Evaluator, Generator, Ingestor — each one independently configurable — pick your own models and thresholds Still local-first. Still privacy-first. Still open source. Still free. Your data never leaves your machine. If you starred the repo last time — pull the new release. If you bookmarked and forgot — now's a good time. Feedback, issues, PRs, or just a ⭐— all welcome. ⭐ 📥 #OpenSource #AgenticAI #AIAgents #GraphRAG #Python #FastAPI #ReactJS #TypeScript #Tauri #JobSearch #CareerTech #PrivacyFirst

Vasu-Devs

121,402 views • 3 months ago

Remember when AI couldn't draw a hand? Seven fingers, knuckles pointing backwards. And the AI spaghetti videos. That was three years ago. Images are done now. Video is close enough that you scrolled past AI ads this week and clocked exactly zero of them. Code writes itself and there are like 40 coding agents. AI voice spent that entire stretch sounding like the lady voice in a 2014 GPS. Flat, evenly spaced and every sentence landing with the same weight, like it's reading from a phone book. Here's why it stayed broken. Bad images are funny. You screenshot the seven fingers, it goes viral for being bad, someone fixes it. Bad audio is just boring. It doesn't fail spectacularly, so it never got that pressure. The bigger problem was the scoring. The whole industry graded AI voices on whether you could make out the words. So the models learned to over pronounce everything, hitting every syllable like a newsreader. Perfectly clear but robotic. Everyone was chasing a score that had nothing to do with sounding human. Meanwhile a small open-source team was doing something harder. Their lead researcher, an ex-NVIDIA engineer, went all in on an approach the rest of the field had written off. Two years early. No funding announcements or launch tour. He just put the whole thing on GitHub for free. It's sitting at 50,000+ stars now. Then they ran the test everyone else avoided. For 10 days they piped real users through their model and every big competitor with the listener never told which was which. Thousands of real people, real scripts. Whichever voice you actually preferred, they logged it. Theirs came out on top. It beat ElevenLabs about 6 times out of 10, head to head. It beat OpenAI's voice model 8 times out of 10. The gap was widest on the breathing, the pauses, the little hesitations, which is exactly the stuff that makes a voice sound like a person instead of a machine reading. They ran on real users rather than a lab, which is more than most of these claims can say. That's Fish Audio. This week they shipped S2.1 Pro: - Clone anyone's voice from 15 seconds of audio - Fast enough to hold a live conversation - 83 languages, one model - Type [whisper] or [sigh] mid-sentence and it does it - Around 70% cheaper than ElevenLabs - Free to download and run yourself Voice was the last thing on the list. Around 20 people with a free repo got there before other billion dollar companies did.

Remember when AI couldn't draw a hand? Seven fingers, knuckles pointing backwards. And the AI spaghetti videos. That was three years ago. Images are done now. Video is close enough that you scrolled past AI ads this week and clocked exactly zero of them. Code writes itself and there are like 40 coding agents. AI voice spent that entire stretch sounding like the lady voice in a 2014 GPS. Flat, evenly spaced and every sentence landing with the same weight, like it's reading from a phone book. Here's why it stayed broken. Bad images are funny. You screenshot the seven fingers, it goes viral for being bad, someone fixes it. Bad audio is just boring. It doesn't fail spectacularly, so it never got that pressure. The bigger problem was the scoring. The whole industry graded AI voices on whether you could make out the words. So the models learned to over pronounce everything, hitting every syllable like a newsreader. Perfectly clear but robotic. Everyone was chasing a score that had nothing to do with sounding human. Meanwhile a small open-source team was doing something harder. Their lead researcher, an ex-NVIDIA engineer, went all in on an approach the rest of the field had written off. Two years early. No funding announcements or launch tour. He just put the whole thing on GitHub for free. It's sitting at 50,000+ stars now. Then they ran the test everyone else avoided. For 10 days they piped real users through their model and every big competitor with the listener never told which was which. Thousands of real people, real scripts. Whichever voice you actually preferred, they logged it. Theirs came out on top. It beat ElevenLabs about 6 times out of 10, head to head. It beat OpenAI's voice model 8 times out of 10. The gap was widest on the breathing, the pauses, the little hesitations, which is exactly the stuff that makes a voice sound like a person instead of a machine reading. They ran on real users rather than a lab, which is more than most of these claims can say. That's Fish Audio. This week they shipped S2.1 Pro: - Clone anyone's voice from 15 seconds of audio - Fast enough to hold a live conversation - 83 languages, one model - Type [whisper] or [sigh] mid-sentence and it does it - Around 70% cheaper than ElevenLabs - Free to download and run yourself Voice was the last thing on the list. Around 20 people with a free repo got there before other billion dollar companies did.

Rez Karim

18,446 views • 1 month ago

Meet WebBrain: An Open-Source, Local-First AI Browser Agent That Reads Pages and Automates Tasks in Chrome and Firefox WebBrain lives inside your browser and can run entirely on your own local model — no cloud, no account, no data leaving your machine. Most "AI browser agents" are a chat box that pastes your page into someone else's server. That's not an agent that lives where you browse — and WebBrain draws a very clear line between the two. It's an open-source (MIT), local-first browser agent for Chrome and Firefox. It runs inside your existing authenticated session, on a model you pick — so with llama.cpp or Ollama, nothing leaves your machine. Here's what's actually interesting: → Two modes, cleanly separated. Ask reads the page (read-only, content scripts). Act clicks and types through the Chrome DevTools Protocol (chrome.debugger) — trusted input events that modern sites honor, reaching cross-origin iframes and shadow DOM. → UI-first by design. For anything that submits, sends, or buys, it drives the visible UI and refuses to hit REST/GraphQL endpoints directly. It starts read-only and asks before consequential actions. → Bring any model. llama.cpp, Ollama, LM Studio, vLLM — or OpenAI, Claude, Gemini, DeepSeek, Groq, OpenRouter. Recommended local: Qwen 3.6 35B (Qwen3.6-35B-A3B), which beat Gemma 4 on the project's screenshot benchmark. → Tuned for cost and privacy. Token-conscious screenshots, oldest-first context trimming, a dedicated vision model, 40+ tools (~20 in Compact mode). No telemetry. No accounts. Full analysis: GitHub Repo: Chrome Extension: Firefox Add-on: Portal:

Meet WebBrain: An Open-Source, Local-First AI Browser Agent That Reads Pages and Automates Tasks in Chrome and Firefox WebBrain lives inside your browser and can run entirely on your own local model — no cloud, no account, no data leaving your machine. Most "AI browser agents" are a chat box that pastes your page into someone else's server. That's not an agent that lives where you browse — and WebBrain draws a very clear line between the two. It's an open-source (MIT), local-first browser agent for Chrome and Firefox. It runs inside your existing authenticated session, on a model you pick — so with llama.cpp or Ollama, nothing leaves your machine. Here's what's actually interesting: → Two modes, cleanly separated. Ask reads the page (read-only, content scripts). Act clicks and types through the Chrome DevTools Protocol (chrome.debugger) — trusted input events that modern sites honor, reaching cross-origin iframes and shadow DOM. → UI-first by design. For anything that submits, sends, or buys, it drives the visible UI and refuses to hit REST/GraphQL endpoints directly. It starts read-only and asks before consequential actions. → Bring any model. llama.cpp, Ollama, LM Studio, vLLM — or OpenAI, Claude, Gemini, DeepSeek, Groq, OpenRouter. Recommended local: Qwen 3.6 35B (Qwen3.6-35B-A3B), which beat Gemma 4 on the project's screenshot benchmark. → Tuned for cost and privacy. Token-conscious screenshots, oldest-first context trimming, a dedicated vision model, 40+ tools (~20 in Compact mode). No telemetry. No accounts. Full analysis: GitHub Repo: Chrome Extension: Firefox Add-on: Portal:

Marktechpost AI

203,011 views • 2 months ago

Apple's product pages have that scroll effect. The camera flies through a 3D world. Products float. Scenes transition. No cuts. One continuous flight driven by your scroll wheel. Here is what one costs to build in 2026. Big storytelling agencies (Noomo, WithLore, Utsubo): $15,000 to $100,000. Three to six months. A team of designers, 3D artists, and WebGL developers. Awwwards-tier flagship from a senior freelancer: $8,000 minimum. Three.js multi-scene site with GSAP scroll: $3,500 to $8,000. That is if you can find a Three.js specialist. They charge $75 to $150 an hour. There are fewer of them than there are brands that want the site. Now meet scroll-world. A free, open-source agent skill for Claude Code and Codex that generates the entire thing from a prompt. Describe your brand. Describe the scenes. "A coffee brand. Start outside the farm. Fly into the roasting facility. Pass through the packaging. Land in the cafe." It generates the isometric stills. It generates the camera flights. It generates the connector clips between scenes from the actual boundary frames of both neighbors, so every seam is pixel-identical. No flicker. No cut. One continuous flight. Then it wires the chain into a portable vanilla JS scroll engine that drops into plain HTML, Next.js, or Vue. One prompt. Hours. Not months. 6,096 stars on GitHub in 25 days. 724 forks. MIT license. Built by cyw, founder and CTO of Hermai AI, out of San Francisco. Here is what scroll-world does: - Interviews you on brand, scenes, art direction, and budget before spending a cent - Generates the isometric stills through GPT Image 2 - Generates the camera flights through Seedance on Monid - Renders a native 9:16 portrait chain for mobile, not a crop - Ships a vanilla JS scrub engine with blob-seek, lazy load, and seam crossfade - Installs into Claude Code as a plugin or Codex through the Vercel skills CLI - Framework-agnostic Here is what scroll-world costs to install: Zero. Forever. Generation burns credits. A 6-scene 1080p chain on Monid pay-per-clip lands around $27, printed before the run starts. Big storytelling agency: $15,000 to $100,000. scroll-world: $27. Awwwards WebGL flagship: $8,000. scroll-world: $27. Three.js contractor for a month: $12,000. scroll-world: this afternoon. Your brand. Your world. Your scroll. 100% Open Source. (Link in the comments)

Nav Toor

359,050 views • 1 month ago

GeoLibre v2.5.0 is here! GeoLibre is a free and open-source, lightweight, cloud-native GIS platform for visualizing, exploring, and analyzing geospatial data. It runs everywhere you do, in the web browser, on the desktop, on mobile, and inside Jupyter notebooks, all while keeping your data local and private. This release is about meeting you where your work already lives: open your QGIS and ArcGIS Pro projects directly, talk to your team on the map itself, explore hyperspectral imagery as a 3D cube, work in discrete global grids, and host sharing and live collaboration on your own server. It is also the most community-driven release so far: 121 pull requests and 52 issues closed in 8 days, from 14 contributors, 12 of whom sent their first contribution to GeoLibre in this cycle. Thank you all. What's new in v2.5.0 - Comments on the map: pin a comment to any location or feature, reply in a thread, resolve it when it is handled. - Live collaboration, on your own server: several people can open the same session and see each other's edits, cursors, and comments in real time. - QGIS and ArcGIS Pro project import: open a .qgs, .qgz, .aprx, or .mapx and get the layers. - Discrete global grids: three new DGGS plugins (A5, DGGRID, and DGGAL) render and identify cells over the current view, plus DGGS Generator, Binning, and Compact processing tools. - Hyperspectral data you can actually read: local NetCDF and HDF grids are colormapped in the browser, a cube gets an RGB band combination picked by wavelength. - Spectral profiles from a click: identify a pixel to chart its spectrum against wavelength. - Smart styling on add: every new layer arrives with its own color and geometry-appropriate sizing, and a Style suggestions strip offers one-click renderers. - Autosave, crash recovery, and project history, plus an Elements panel for managing every annotation from a list, a Layer Library for saving a fully configured layer and re-adding it to any later project, and project duplication and templates. - GeoLibre Desktop is on the Mac App Store, and Thai brings the shipped locales to 16 languages besides English, all at 100% coverage. Try it out - Launch GeoLibre Web: - GitHub: - Documentation: - Release notes: #GIS #Geospatial #OpenSource #RemoteSensing #DataVisualization #MapLibre #GeoLibre

GeoLibre v2.5.0 is here! GeoLibre is a free and open-source, lightweight, cloud-native GIS platform for visualizing, exploring, and analyzing geospatial data. It runs everywhere you do, in the web browser, on the desktop, on mobile, and inside Jupyter notebooks, all while keeping your data local and private. This release is about meeting you where your work already lives: open your QGIS and ArcGIS Pro projects directly, talk to your team on the map itself, explore hyperspectral imagery as a 3D cube, work in discrete global grids, and host sharing and live collaboration on your own server. It is also the most community-driven release so far: 121 pull requests and 52 issues closed in 8 days, from 14 contributors, 12 of whom sent their first contribution to GeoLibre in this cycle. Thank you all. What's new in v2.5.0 - Comments on the map: pin a comment to any location or feature, reply in a thread, resolve it when it is handled. - Live collaboration, on your own server: several people can open the same session and see each other's edits, cursors, and comments in real time. - QGIS and ArcGIS Pro project import: open a .qgs, .qgz, .aprx, or .mapx and get the layers. - Discrete global grids: three new DGGS plugins (A5, DGGRID, and DGGAL) render and identify cells over the current view, plus DGGS Generator, Binning, and Compact processing tools. - Hyperspectral data you can actually read: local NetCDF and HDF grids are colormapped in the browser, a cube gets an RGB band combination picked by wavelength. - Spectral profiles from a click: identify a pixel to chart its spectrum against wavelength. - Smart styling on add: every new layer arrives with its own color and geometry-appropriate sizing, and a Style suggestions strip offers one-click renderers. - Autosave, crash recovery, and project history, plus an Elements panel for managing every annotation from a list, a Layer Library for saving a fully configured layer and re-adding it to any later project, and project duplication and templates. - GeoLibre Desktop is on the Mac App Store, and Thai brings the shipped locales to 16 languages besides English, all at 100% coverage. Try it out - Launch GeoLibre Web: - GitHub: - Documentation: - Release notes: #GIS #Geospatial #OpenSource #RemoteSensing #DataVisualization #MapLibre #GeoLibre

Qiusheng Wu

26,788 views • 1 month ago

SOME GUY BUILT A 24/7 ONLINE RAVE IN YOUR BROWSER AND IT IS COMPLETELY FREE TO ATTEND it's called Hallucinate. you click one link, your browser loads a 3D field at night, and suddenly you are standing in a crowd of hundreds of low-poly strangers dancing under a moon to a youtube video playing on a giant screen. there is no login. there is no wallet. there is no signup. you are just at the rave. the rave has its own keyboard. > 7 and 8 cycle your hair color > 9 and 0 cycle your skin tone > U and I cycle your hair style > L and ; cycle your dance moves > V is the wave emote > B is the bounce emote > SPACE bar is literally labeled "speak" every other person in the room is a real human pressing the same keys, in real time. nobody ever stops dancing because the creator wrote on hacker news: "everybody dancing forever is crucial for the feeling. you don't stop dancing in a rave." the dev who made it goes by the handle stagas. he says the entire thing was "vibe coded in its entirety" by an LLM. the commit messages on the repo are three words long: "cool." "jump jump." "perf." he has about 500 other repos on his github account. the post went up on Show HN two days ago. it is currently sitting at 435 points and 193 comments and the top discussion in the thread is people arguing whether magic numbers in shaders.ts disqualify you from getting a job. we built the global internet so that two strangers across the planet could fall in love or trade stocks or topple regimes. and the very best thing happening on it right now is a thousand cubes dancing forever under a fake moon. the internet is healing.

SOME GUY BUILT A 24/7 ONLINE RAVE IN YOUR BROWSER AND IT IS COMPLETELY FREE TO ATTEND it's called Hallucinate. you click one link, your browser loads a 3D field at night, and suddenly you are standing in a crowd of hundreds of low-poly strangers dancing under a moon to a youtube video playing on a giant screen. there is no login. there is no wallet. there is no signup. you are just at the rave. the rave has its own keyboard. > 7 and 8 cycle your hair color > 9 and 0 cycle your skin tone > U and I cycle your hair style > L and ; cycle your dance moves > V is the wave emote > B is the bounce emote > SPACE bar is literally labeled "speak" every other person in the room is a real human pressing the same keys, in real time. nobody ever stops dancing because the creator wrote on hacker news: "everybody dancing forever is crucial for the feeling. you don't stop dancing in a rave." the dev who made it goes by the handle stagas. he says the entire thing was "vibe coded in its entirety" by an LLM. the commit messages on the repo are three words long: "cool." "jump jump." "perf." he has about 500 other repos on his github account. the post went up on Show HN two days ago. it is currently sitting at 435 points and 193 comments and the top discussion in the thread is people arguing whether magic numbers in shaders.ts disqualify you from getting a job. we built the global internet so that two strangers across the planet could fall in love or trade stocks or topple regimes. and the very best thing happening on it right now is a thousand cubes dancing forever under a fake moon. the internet is healing.

Nav Toor

249,078 views • 3 months ago

While the world doomscrolls 15-second TikToks and loses its attention span.. YOU SHOULD CHECK OUT THIS NEW REPO A Chinese college kid built MiroFish in just 10 days, scored $4M funding and ByteDance just dropped the upgrade that turns it into a prediction monster. Fourth-year student Guo Hanjiang vibe coded MiroFish: thousands of autonomous AI agents simulating entire societies in real time. See the details in the post below. Feed it any news, report, or historical data - watch markets, crowds, and politics react exactly as they would. GitHub went nuclear, he scored $4M funding from Shanda Group, and Polymarket traders are already printing +$120k+ using his SPX and event simulations. But long runs had one fatal flaw: agents got amnesia, hallucinations, and context overload. Classic RAG garbage. Now the same company behind TikTok (via VolcEngine) just open-sourced the fix: OpenViking - already at 11.6k+ stars on GitHub. Repo: It turns chaotic memory into a clean, structured filesystem: -> viking://user/memories/ (your habits + past outcomes) -> viking://agent/skills/ (trading and analysis superpowers) Smart 3-layer hierarchy: .L0 - 100-token ultra-summary .L1 - quick overview .L2 - full details (opened only when needed) Agents browse folders intelligently instead of dumping everything. Result: > No more forgetting crucial facts from the start of a simulation > Way fewer hallucinations > Massive API token savings > Self-updating memory - agents get smarter after every run MiroFish + OpenViking = absolute nuclear edge for Polymarket and event prediction. I will use it for my private bot. Thousands of agents now run with perfect long-term memory, stay sharp for 100+ steps, and deliver hyper-accurate probabilities. This combo is about to change the game for anyone trading predictions. The irony is insane: the company that killed human attention with short videos just gave AI agents eternal, structured, self-evolving memory. Who’s already running this combo on Polymarket? Save this. The real alpha just dropped.

While the world doomscrolls 15-second TikToks and loses its attention span.. YOU SHOULD CHECK OUT THIS NEW REPO A Chinese college kid built MiroFish in just 10 days, scored $4M funding and ByteDance just dropped the upgrade that turns it into a prediction monster. Fourth-year student Guo Hanjiang vibe coded MiroFish: thousands of autonomous AI agents simulating entire societies in real time. See the details in the post below. Feed it any news, report, or historical data - watch markets, crowds, and politics react exactly as they would. GitHub went nuclear, he scored $4M funding from Shanda Group, and Polymarket traders are already printing +$120k+ using his SPX and event simulations. But long runs had one fatal flaw: agents got amnesia, hallucinations, and context overload. Classic RAG garbage. Now the same company behind TikTok (via VolcEngine) just open-sourced the fix: OpenViking - already at 11.6k+ stars on GitHub. Repo: It turns chaotic memory into a clean, structured filesystem: -> viking://user/memories/ (your habits + past outcomes) -> viking://agent/skills/ (trading and analysis superpowers) Smart 3-layer hierarchy: .L0 - 100-token ultra-summary .L1 - quick overview .L2 - full details (opened only when needed) Agents browse folders intelligently instead of dumping everything. Result: > No more forgetting crucial facts from the start of a simulation > Way fewer hallucinations > Massive API token savings > Self-updating memory - agents get smarter after every run MiroFish + OpenViking = absolute nuclear edge for Polymarket and event prediction. I will use it for my private bot. Thousands of agents now run with perfect long-term memory, stay sharp for 100+ steps, and deliver hyper-accurate probabilities. This combo is about to change the game for anyone trading predictions. The irony is insane: the company that killed human attention with short videos just gave AI agents eternal, structured, self-evolving memory. Who’s already running this combo on Polymarket? Save this. The real alpha just dropped.

slash1s

293,591 views • 5 months ago

I just closed a $70,000 deal with help from 10 Hermes agents Here is how it happened: A couple of months ago, one of my clients asked me to review their IT spending. The biggest surprise was Atlassian. They were using Jira, Confluence, Jira Service Management, and several paid plugins because the built in reporting and time tracking were not meeting their needs. With more than 300 users, they were spending around $140,000 every year. They had been paying that for six years. That is roughly $840,000. So I asked them: “If I build something that replaces most of these tools, improves the workflows your team struggles with, and cuts your bill in half, would you pay me $70,000 per year?” They said yes. So I put 10 Hermes agents to work For the next month, they worked alongside me for almost 8 hours a day. About half of the code was built using Fable 5 and the other half using Opus 4.8. I spent approximately $11,000 on tokens :P That sounds expensive until you compare it with a $70,000 yearly contract and a product I can now offer to other companies. One month later, the product was ready. The client loved it and gave me the green light to open it to the public and it's my code :) I called it Kapvel Kapvel includes: • Project management and Jira ticket imports • A full service desk with a customer portal • Built in time tracking and reporting • Documents with real time collaboration • Spreadsheets, presentations, and whiteboards • Electronic signatures and approvals • QA test case management • A dedicated UAT portal • GitHub, Bitbucket, and GitLab integrations • Pull request workflows inside the same platform • Custom themes and branded portals The part I am most excited about is what happens when someone submits a support ticket to me: My Hermes agents can immediately review the request, investigate the issue, and start working on a solution. Kapvel is basically: Jira + Confluence + DocuSign + Everhour + QA + UAT All inside one platform, with simpler pricing and fewer third party plugins. This is not a mockup The client is already using it Now I am preparing to open Kapvel to other teams Comment KAPVEL for early access.

I just closed a $70,000 deal with help from 10 Hermes agents Here is how it happened: A couple of months ago, one of my clients asked me to review their IT spending. The biggest surprise was Atlassian. They were using Jira, Confluence, Jira Service Management, and several paid plugins because the built in reporting and time tracking were not meeting their needs. With more than 300 users, they were spending around $140,000 every year. They had been paying that for six years. That is roughly $840,000. So I asked them: “If I build something that replaces most of these tools, improves the workflows your team struggles with, and cuts your bill in half, would you pay me $70,000 per year?” They said yes. So I put 10 Hermes agents to work For the next month, they worked alongside me for almost 8 hours a day. About half of the code was built using Fable 5 and the other half using Opus 4.8. I spent approximately $11,000 on tokens :P That sounds expensive until you compare it with a $70,000 yearly contract and a product I can now offer to other companies. One month later, the product was ready. The client loved it and gave me the green light to open it to the public and it's my code :) I called it Kapvel Kapvel includes: • Project management and Jira ticket imports • A full service desk with a customer portal • Built in time tracking and reporting • Documents with real time collaboration • Spreadsheets, presentations, and whiteboards • Electronic signatures and approvals • QA test case management • A dedicated UAT portal • GitHub, Bitbucket, and GitLab integrations • Pull request workflows inside the same platform • Custom themes and branded portals The part I am most excited about is what happens when someone submits a support ticket to me: My Hermes agents can immediately review the request, investigate the issue, and start working on a solution. Kapvel is basically: Jira + Confluence + DocuSign + Everhour + QA + UAT All inside one platform, with simpler pricing and fewer third party plugins. This is not a mockup The client is already using it Now I am preparing to open Kapvel to other teams Comment KAPVEL for early access.

Luke The Dev

25,493 views • 1 month ago

Andrej Karpathy said: "There's room for an incredible new product in the second brain space" This might be it. (bookmark it) Everyone is suddenly building a second brain. Karpathy's LLM wiki pattern went viral, and half of X is now hand-wiring Obsidian to Claude Code so an agent maintains their notes for them. The idea is beautiful: stop making your AI re-read raw notes on every question. Let it build a wiki that compounds. As Karpathy put it, "LLMs don't get bored, they don't forget to update a cross-reference (backlinks), and can touch 15 files in one pass." But if you start doing it manually, it becomes a project in itself. You wire up the vault, the agents, the schedules, the integrations, and then you babysit all of it. So I sat down with Arjun, who actually built the open source version of this, and we broke down what it looks like when the whole thing already works out of the box. It just crossed 15K stars on GitHub. Think Claude's desktop app, open source, with two things layered on top: → A work brain: background agents index your emails, meetings, and notes into a living knowledge graph that updates itself as you work. → Work surfaces: chat is not the best interface for real work, so you get an email client, a meeting note taker, a browser, and a code mode where you and the AI actually collaborate. The part that got me: a customer email comes in asking for a product change, a background agent triages it, spins up Claude Code in its own worktree, and the feature is written before you are back at your desk. Bring your existing Obsidian vault, connect Slack, X, and Fireflies, and let it run your day. Here's the full breakdown of what we covered in this session: Enjoy! 00:00 Intro 01:08 What is Roboat (an open source AI co-worker) 02:42 The second brain (a knowledge graph of your work) 04:01 Bringing your existing Obsidian vault in 04:46 Work surfaces 05:29 Meetings and automatic note taking 06:53 Connecting Slack, X and other sources 07:55 Background agents that run your day 09:24 Code mode (Claude Code and Codex) 10:18 Demo: from an email to written code 14:28 Guardrails: approvals and agent workspaces 17:15 Scheduling agents on a cron 18:52 The browser work surface (browser use) 20:42 Wrapping up: automating your whole day 22:44 Outro Checkout Rowboat's GitHub repo: (don't forget to star 🌟) My co-founder recently wrote a great article on the same idea, and I highly recommend reading it as well. The article is quoted below. Here's my session with Arjun:

Andrej Karpathy said: "There's room for an incredible new product in the second brain space" This might be it. (bookmark it) Everyone is suddenly building a second brain. Karpathy's LLM wiki pattern went viral, and half of X is now hand-wiring Obsidian to Claude Code so an agent maintains their notes for them. The idea is beautiful: stop making your AI re-read raw notes on every question. Let it build a wiki that compounds. As Karpathy put it, "LLMs don't get bored, they don't forget to update a cross-reference (backlinks), and can touch 15 files in one pass." But if you start doing it manually, it becomes a project in itself. You wire up the vault, the agents, the schedules, the integrations, and then you babysit all of it. So I sat down with Arjun, who actually built the open source version of this, and we broke down what it looks like when the whole thing already works out of the box. It just crossed 15K stars on GitHub. Think Claude's desktop app, open source, with two things layered on top: → A work brain: background agents index your emails, meetings, and notes into a living knowledge graph that updates itself as you work. → Work surfaces: chat is not the best interface for real work, so you get an email client, a meeting note taker, a browser, and a code mode where you and the AI actually collaborate. The part that got me: a customer email comes in asking for a product change, a background agent triages it, spins up Claude Code in its own worktree, and the feature is written before you are back at your desk. Bring your existing Obsidian vault, connect Slack, X, and Fireflies, and let it run your day. Here's the full breakdown of what we covered in this session: Enjoy! 00:00 Intro 01:08 What is Roboat (an open source AI co-worker) 02:42 The second brain (a knowledge graph of your work) 04:01 Bringing your existing Obsidian vault in 04:46 Work surfaces 05:29 Meetings and automatic note taking 06:53 Connecting Slack, X and other sources 07:55 Background agents that run your day 09:24 Code mode (Claude Code and Codex) 10:18 Demo: from an email to written code 14:28 Guardrails: approvals and agent workspaces 17:15 Scheduling agents on a cron 18:52 The browser work surface (browser use) 20:42 Wrapping up: automating your whole day 22:44 Outro Checkout Rowboat's GitHub repo: (don't forget to star 🌟) My co-founder recently wrote a great article on the same idea, and I highly recommend reading it as well. The article is quoted below. Here's my session with Arjun:

Akshay 🚀

45,815 views • 2 months ago

Massive breakthrough here! Someone fixed every major flaw in Jupyter Notebooks. The .ipynb format is stuck in 2014. It was built for a different era - no cloud collaboration, no AI agents, no team workflows. Change one cell, and you get 50+ lines of JSON metadata in your git diff. Code reviews become a nightmare. Want to share a database connection across notebooks? Configure it separately in each one. Need comments or permissions? Too bad. Jupyter works for solo analysis but breaks for teams building production AI systems. Deepnote just open-sourced the solution (Apache 2.0 license) They've built a new notebook standard that actually fits modern workflows: ↳ Human-readable YAML - Git diffs show actual code changes, not JSON noise. Code reviews finally work. ↳ Project-based structure - Multiple notebooks share integrations, secrets, and environment settings. Configure once, use everywhere. ↳ 23 new block - SQL, interactive inputs, charts, and KPIs as first-class citizens. Build data apps, not just analytics notebooks. ↳ Multi-language support - Python and SQL in one notebook. Modern data work isn't single-language anymore. ↳ Full backward and forward compatibility: convert any Jupyter notebook to Deepnote and vice versa with one command. npx @ deepnote/convert notebook.ipynb Then open it in VS Code, Cursor, WindSurf, or Antigravity. Your existing notebooks migrate instantly. Their cloud version adds real-time collaboration with comments, permissions, and live editing. I've shared the GitHub repo link in the replies! It's 100% open-source.

Massive breakthrough here! Someone fixed every major flaw in Jupyter Notebooks. The .ipynb format is stuck in 2014. It was built for a different era - no cloud collaboration, no AI agents, no team workflows. Change one cell, and you get 50+ lines of JSON metadata in your git diff. Code reviews become a nightmare. Want to share a database connection across notebooks? Configure it separately in each one. Need comments or permissions? Too bad. Jupyter works for solo analysis but breaks for teams building production AI systems. Deepnote just open-sourced the solution (Apache 2.0 license) They've built a new notebook standard that actually fits modern workflows: ↳ Human-readable YAML - Git diffs show actual code changes, not JSON noise. Code reviews finally work. ↳ Project-based structure - Multiple notebooks share integrations, secrets, and environment settings. Configure once, use everywhere. ↳ 23 new block - SQL, interactive inputs, charts, and KPIs as first-class citizens. Build data apps, not just analytics notebooks. ↳ Multi-language support - Python and SQL in one notebook. Modern data work isn't single-language anymore. ↳ Full backward and forward compatibility: convert any Jupyter notebook to Deepnote and vice versa with one command. npx @ deepnote/convert notebook.ipynb Then open it in VS Code, Cursor, WindSurf, or Antigravity. Your existing notebooks migrate instantly. Their cloud version adds real-time collaboration with comments, permissions, and live editing. I've shared the GitHub repo link in the replies! It's 100% open-source.

Akshay 🚀

33,358 views • 9 months ago

I BUILT A BOT THAT PREDICTS NBA GAMES BEFORE TIPOFF. IT RUNS EVERY DAY AT 5:00 PM. HERE'S WHAT'S INSIDE. Three data sources. Three probability layers. One automated pipeline. No manual input. Every evening the bot wakes up. Pulls today's NBA slate. Builds features from live data.Generates predictions. Sends them to me before the first whistle. The ML layer: XGBoost on three seasons of data. Four Factors. ELO ratings. Net rating. Pace. Back-to-back fatigue. Rest days. Travel distance. Every stat computed only from data available before the game. No future leakage. Ever. The sportsbook layer: spread and moneyline converted into implied probabilities. The spread alone contains the aggregated intelligence of an entire betting market. The bot doesn't fight it. It absorbs it. The Polymarket layer: live crowd-sourced probabilities from a decentralized prediction market. No vig. Prices shift in seconds when a star sits out. The bot pulls them through the Gamma API in real time. Then all three layers talk to each other. All three agree - high confidence bet. Polymarket diverges from the books - someone knows something. My model diverges from both - either I found edge or I found noise. The bot flags every disagreement automatically. The fourth layer: Claude API reads everything. Injury reports. Schedule context. Matchup history. Generates a written explanation for every single prediction. No black box. Every call has reasoning attached. The stack: Python. nba_api. pandas. XGBoost. scikit-learn. SHAP. Polymarket Gamma API. Claude API. Flask. GitHub Actions. It took mass of work to build. It takes zero effort to run. 5:00 PM ET. Every day. Predictions land before tipoff. Trading here: I didn't build a model. I built a machine.

I BUILT A BOT THAT PREDICTS NBA GAMES BEFORE TIPOFF. IT RUNS EVERY DAY AT 5:00 PM. HERE'S WHAT'S INSIDE. Three data sources. Three probability layers. One automated pipeline. No manual input. Every evening the bot wakes up. Pulls today's NBA slate. Builds features from live data.Generates predictions. Sends them to me before the first whistle. The ML layer: XGBoost on three seasons of data. Four Factors. ELO ratings. Net rating. Pace. Back-to-back fatigue. Rest days. Travel distance. Every stat computed only from data available before the game. No future leakage. Ever. The sportsbook layer: spread and moneyline converted into implied probabilities. The spread alone contains the aggregated intelligence of an entire betting market. The bot doesn't fight it. It absorbs it. The Polymarket layer: live crowd-sourced probabilities from a decentralized prediction market. No vig. Prices shift in seconds when a star sits out. The bot pulls them through the Gamma API in real time. Then all three layers talk to each other. All three agree - high confidence bet. Polymarket diverges from the books - someone knows something. My model diverges from both - either I found edge or I found noise. The bot flags every disagreement automatically. The fourth layer: Claude API reads everything. Injury reports. Schedule context. Matchup history. Generates a written explanation for every single prediction. No black box. Every call has reasoning attached. The stack: Python. nba_api. pandas. XGBoost. scikit-learn. SHAP. Polymarket Gamma API. Claude API. Flask. GitHub Actions. It took mass of work to build. It takes zero effort to run. 5:00 PM ET. Every day. Predictions land before tipoff. Trading here: I didn't build a model. I built a machine.

zostaff

48,649 views • 4 months ago

A once-in-a-lifetime experience – Paris Olympics 2024. From the time I was a kid, eagerly flipping to the sports section as soon as the newspaper arrived, the Olympics has always been the event I dreamed of attending. It’s not just any event; it’s the event. If someone asked me about my life goals, I’d say attending every Olympics until my last breath. But getting to Paris wasn’t easy. From facing a visa rejection to this being my first international trip—and solo at that—there were challenges. I wasn’t too comfortable in front of the camera either, but I decided to give it a shot. I hope you all enjoyed the content I created along the way. Your support and encouragement have meant the world to me. The Olympics is truly a vibe. The energy is electric, with everyone in high spirits, cheering on the world’s best athletes. I spent my days hopping from one arena to another, navigating the city by train, meeting amazing people from different countries, and making new friends along the way. This experience is one I'll carry with me forever. This trip also made me realize that I could have spent the last decade chasing more of my dreams, but it’s better late than never. Here's to more such experiences. Au revoir, Paris. You have all my heart. 💙 Sportwalk Media Sportwalk Studio The Bharat Army #Sportwalk #ParisOlympics #Paris2024 #OlympicGames #Olympics #indiaatolympics #TeamIndia #Cheer4Bharat #Cheer4India #paris

A once-in-a-lifetime experience – Paris Olympics 2024. From the time I was a kid, eagerly flipping to the sports section as soon as the newspaper arrived, the Olympics has always been the event I dreamed of attending. It’s not just any event; it’s the event. If someone asked me about my life goals, I’d say attending every Olympics until my last breath. But getting to Paris wasn’t easy. From facing a visa rejection to this being my first international trip—and solo at that—there were challenges. I wasn’t too comfortable in front of the camera either, but I decided to give it a shot. I hope you all enjoyed the content I created along the way. Your support and encouragement have meant the world to me. The Olympics is truly a vibe. The energy is electric, with everyone in high spirits, cheering on the world’s best athletes. I spent my days hopping from one arena to another, navigating the city by train, meeting amazing people from different countries, and making new friends along the way. This experience is one I'll carry with me forever. This trip also made me realize that I could have spent the last decade chasing more of my dreams, but it’s better late than never. Here's to more such experiences. Au revoir, Paris. You have all my heart. 💙 Sportwalk Media Sportwalk Studio The Bharat Army #Sportwalk #ParisOlympics #Paris2024 #OlympicGames #Olympics #indiaatolympics #TeamIndia #Cheer4Bharat #Cheer4India #paris

Divakar KS

16,364 views • 2 years ago