Five components that together cover the full loop a research engineer owns at a systematic trading firm — from a quant's first hypothesis in a notebook to a live signal feeding the production trading stack, with the lineage to audit any decision in between.
The framework you spend most of your day in. Declarative model definitions, walk-forward CV, leakage-safe feature evaluation, and a benchmark grid that lets a quant compare three weeks of model variants over coffee.
Treat features like a build system: declarative dependencies, point-in-time correctness baked in, content-addressed caching, and the same graph at training time and serving time.
The production half of the loop: the same feature graph that ran in research, now consuming a market-data stream and emitting signals into the trading stack — with point-in-time guarantees, lineage, and a kill switch.
A self-hosted model registry that's the source of truth for what trains, what ships, and what's running. Each model carries its full lineage — code commit, feature graph hash, data window, owner, evaluation report — and the production platform refuses to load anything it didn't register.
A small library and a sidecar that let a researcher prototype freely in a notebook, then promote the cell that worked into a versioned, reproducible pipeline — without rewriting it. The notebook is the IDE; the platform is the runtime.