I will build a multi agent reinforcement learning algorithm

I build custom reinforcement learning environments and multi-agent simulations for games, robotics, logistics, and research applications.

Recent work includes a competitive multi-agent football simulator (curriculum learning from 1v1 to 11v11, self-play PPO training) and a full Chinese Mahjong environment implementing 77 of 81 official scoring rules plus a multi-agent transit-network simulation built on real city transit data, from the raw data pipeline through a trained policy with measured performance against a baseline.

What you get is a tested, documented codebase, not a notebook: a proper Gymnasium/PettingZoo-compatible environment, an RLlib training pipeline, and a written evaluation showing your trained agent's performance against a baseline. If it doesn't beat baseline, it's not done.

Before we start, I'll confirm reinforcement learning is actually the right approach for your problem some "AI agent" tasks are better and more cheaply solved with simpler methods, and I'll tell you if that's the case rather than take the job anyway.