overmind-core/overmind ? reverse-engineered prompt

Build me a self hosted platform for improving AI agents from their production traces. I want a simple web console where I can connect an agent repo, see its capabilities, prompts, tools, tasks, and traces, then turn those traces into versioned datasets for evals and fine tuning. It should let me score runs, compare model versions, run backtests against saved traces, and deploy a trained model behind one easy API. Also make it work with a coding agent through a command style workflow, so I can initialize a repo, sync it, and run things like tracing, dataset creation, finetuning, and optimisation from the same chat.

Please wire up the full local experience with Docker Compose, a backend API, a frontend dashboard, background workers, and basic observability. If you need to, look up the current docs for any third party services before implementing.