Problem
Experiment workflows tend to sprawl. Configuration lives in one place, run metadata in another, and evaluation results in a third. This dummy project stands in for a tool that gives a single view of experiments, metrics, and artifacts.
System
The system collects experiment metadata, stores run history, and exposes a small web UI for comparing runs and tracing results back to code and config. The point of the page is to show the level of detail I want for real project write-ups.
Implementation notes
- FastAPI service for run metadata and experiment APIs
- Postgres for structured run storage and queryable comparisons
- Background workers for metric ingestion and artifact indexing
- Small frontend for run comparison, filtering, and evaluation views