At Sonoma Raceway, Google Developer Experts built an AI Race Coach using Google Antigravity to bridge software code and vehicle telemetry. Tapping 100+ car sensors on a Pixel 10, Gemma 4 ran locally to deliver real-time audio coaching.

Learn more about this project here:

When 3.5 pro?

Hear me out. 🤣

What an awesome project, great job & congrats to all the devs 💪 😎 I hope it evolves with Nürburgring as a next step, with a very good driver that hasn't been there yet. That'd be an awesome benchmark 🏁 🤩

I wanted to measure Gemma 4 performance on my vehicles, how well, how fast, how long, any question answered on any Gemma 4 model you can fit on your macOS or iOS. All built entirely with @antigravity , and so much fun!

That's awesome, seeing AI actually get handson with real cars like that. Bet the drivers loved having a coach that never gets tired or distracted.

This is awesome

now THIS is stunning use of Gemma 4. A breakdown of how the audio coaching component was built would be epic.

@NathenFernandes gemma 4 can do this and you're worried about tool-calling

If only google antigravity could be more competitive in a societal sense; Google's models are profoundly lobotomized on a consumer/commercial level. In a restrictive/scientific environment, I suppose the physics/workflows are conducive, but sadly, the bureaucracy eats efficiency.

Cool but extremely misleading to call this ‘AI race coach’. model merely acts as a constrained rule selector and language layer, not as the driving analyst, which is done by heuristics!

How many parameters does this Gemma 4 have?

Was it @mlnomadpy on the car?

I am very curious to know like which coding model does Google is using internally for their work?

finaly, a real case , real dev, not just a benchmark or ranking

Thanks! The most interesting part is how quickly individuals can now achieve what required teams before.

So there is one more job category gone

On-device processing could also reduce latency and keep sensitive vehicle data private during high-speed sessions.

You are really up to winning this AI race. This is a great leap forward into making significant impact in areas where AI seems to be slow (in reference to adoption).

keep innovating! keep automating! keep resonance! 🐾💙🎉 @barkmeta @GodsBurnt @Shieldmetax Advertise on X's biggest media group, CSN:

That's a wild use case for on-device AI. Telemetry to audio coaching in real time sounds like a game changer for the track.

Bridging code to live telemetry via Antigravity — clean concept. How did you handle sync between sensor rates?

Can we please get all the AI labs into racing?

gemma 4 whispering in your ear while you miss the apex feels like cheating. hope they don't make it too good or we'll all just drive slow

When opus 5? We are not paying for your outdated model and your stupid Gemini

how long until gemma is just driving the car?

is this like having Lewis Hamilton by your side?

with 100+ sensors, timestamp drift will break coaching before model quality does

Gemma running locally on a Pixel for real time coaching is wild.

把遥测直接塞进本地模型，赛道上真挺敢玩

I once took my Dad's Lexus, of course without telling him, to the Sonoma raceway on an open day! No I did not win but enjoyed the goose bumps!

Scaling Law 的边际效益正在递减。当所有人都可以烧钱刷榜时，真正的护城河在于谁能率先把技术转化为用户离不开的产品。这条路没有捷径，也没有护城河可以永远不被复制。

IYKYK