A three-application finance suite sharing one foundation: CopilotQuant turns a plain-English strategy description into a typed specification and backtests it, RiskLens analyses portfolio risk, and FilingMind reads company filings. Built from adversarially-reviewed specifications rather than improvised.
The problem
Most backtesting tools force you to write code before you can test an idea, and most give you a different answer if you run them twice. Neither is acceptable if the number is going to inform a decision.
What I built
All three units built and verified end to end in the browser: login, parse a strategy from English, save it, run the backtest through the live Python service, and read the metrics, equity curve and fill blotter. Reproducibility proven by two runs returning identical figures to six decimal places.
How it works
A monorepo splits the interface from the computation: a TypeScript web application handles parsing, storage and presentation, while a Python service owns the actual backtest engine, written in plain pandas and numpy so the maths stays auditable. Natural language is parsed into a typed strategy specification — a validated object, not a prompt — which is what actually gets executed. Determinism is enforced by keying the data generator off a hash of the seed and symbol, so the same specification produces a byte-identical result run after run.
Highlights
- English strategy description parsed into a typed, validated specification before execution
- Backtest engine in pure pandas/numpy so every calculation is auditable
- Byte-identical results across repeated runs — determinism proven, not assumed
- Dashboard with metrics, equity-vs-benchmark, drawdown and a full fill blotter
- Grounded explanations of results generated from the actual run data
- Built from adversarially-reviewed specs with pinned architectural decisions
Let's build yours.
Tell me what you're trying to ship — you'll get a scoped plan and a straight answer.