yazin

@yazins • 8,073 subscribers

messing with AI / founder @amalinvest (YC, W22)

Shorts

Introducing: OpenGranola 🔥 I built an open source meeting copilot for macOS. It transcribes both sides of your call on-device, searches your own notes in real time, and hands you talking points right when the conversation needs them. No audio leaves your Mac. Point it at a folder of markdown files, pick any LLM through OpenRouter (Claude, GPT-4o, Gemini, Llama), and it just works. It's invisible to screen share too — nobody knows you have it. The whole thing is open source. Link below

Introducing: OpenGranola 🔥 I built an open source meeting copilot for macOS. It transcribes both sides of your call on-device, searches your own notes in real time, and hands you talking points right when the conversation needs them. No audio leaves your Mac. Point it at a folder of markdown files, pick any LLM through OpenRouter (Claude, GPT-4o, Gemini, Llama), and it just works. It's invisible to screen share too — nobody knows you have it. The whole thing is open source. Link below

292,712 Aufrufe

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

BREAKTHROUGH on offline-tarteel today Cyberistic hit 100% recall, 100% precision, and 100% sequence accuracy, with 0.84s average CPU latency. I can't believe i'm writing this sentence! His idea was simple: instead of trying to "understand everything" in one shot, it listens to the recitation, finds the Qur’an verses that sound most likely, then checks the small audio clues against those verses until the right one wins. That's what made it fast enough to run offline, and accurate enough to feel real. This is the part that makes me happiest: open source worked exactly the way I hoped! I put my earlier work out there. Someone else picked it up, pushed it way further, and now we have something I genuinely didn’t think we’d reach this soon. link below!

BREAKTHROUGH on offline-tarteel today Cyberistic hit 100% recall, 100% precision, and 100% sequence accuracy, with 0.84s average CPU latency. I can't believe i'm writing this sentence! His idea was simple: instead of trying to "understand everything" in one shot, it listens to the recitation, finds the Qur’an verses that sound most likely, then checks the small audio clues against those verses until the right one wins. That's what made it fast enough to run offline, and accurate enough to feel real. This is the part that makes me happiest: open source worked exactly the way I hoped! I put my earlier work out there. Someone else picked it up, pushed it way further, and now we have something I genuinely didn’t think we’d reach this soon. link below!

29,445 Aufrufe • vor 24 Tagen

BREAKTHROUGH 85% accuracy at just 115MB (!) i almost can't believe it 📹 demo video below for context: i spent days on offline Quran recognition. Whisper, Moonshine, wav2vec2 fine-tunes, layer pruning, CTC rescoring, beam search tricks. nothing cracked 81% accuracy. i then plugged in NVIDIA's pretrained FastConformer and immediately got 90.7%. also, it's 10x faster 🐇 (0.33s vs 3.24s) and 10x smaller 👌 (115MB vs 1.2GB). i tried fine-tuning it on Quranic audio, figuring I could push it even higher. Ran four different configs. every single one came back worse than the stock pretrained model. no idea what's going on there yet. code/repo in next tweet

BREAKTHROUGH 85% accuracy at just 115MB (!) i almost can't believe it 📹 demo video below for context: i spent days on offline Quran recognition. Whisper, Moonshine, wav2vec2 fine-tunes, layer pruning, CTC rescoring, beam search tricks. nothing cracked 81% accuracy. i then plugged in NVIDIA's pretrained FastConformer and immediately got 90.7%. also, it's 10x faster 🐇 (0.33s vs 3.24s) and 10x smaller 👌 (115MB vs 1.2GB). i tried fine-tuning it on Quranic audio, figuring I could push it even higher. Ran four different configs. every single one came back worse than the stock pretrained model. no idea what's going on there yet. code/repo in next tweet

62,326 Aufrufe • vor 4 Monaten

Keine weiteren Inhalte verfügbar