in the past couple of months or so I've built and completed 5 really cool projects... trust me this is worth reading... all of the details can be found on my website, but here’s an overview
> Formula One Reinforcement Learning - Built a custom F1 physics simulator, calibrated it to real F1 telemety and then ran evolutionary search, but more importantly trained n SAC agent on top of it to match real F1 timings.
> MonopolyBench - Built a custom multi-agent benchmark to have LLMs play monopoly against each other. Long horizon economic agents! (who do you think plays best and comes out on top?)
> Web Crawler in Go - Mostly to learn Go, but achieved some really good benchmarks speeds due to Goroutines and channels. Really cool and fun to learn about systems and the internet in general. (no pic bcuz well yk)
> BeyondChat - A custom LLM powered workspace that escapes the generic chat based interface and shifts everything to custom agents and studios (Writing, Research, Finance, Data, Image, etc.). Everything connects with Google and Notion and is built around organizations for efficient collaboration.
> NovelBench - Are LLMs capable of producing truly novel ideas and drafting complete plans? Built with another custom harness and a graph based scoring mechanism. (novelbench(.)vercel(.)app)
It's been busy but it's also been extremely fun. I've learnt so much by completing and deploying these projects end to end. There's so much more to come...