← All projects

★ Featured · 2025 – present · Active

Automated Trading Research Platform

Personal research project

A personal platform for building and paper-trading systematic strategies, designed so that research results can be trusted: the architecture makes common sources of silent error — mutated history, bypassed risk checks, lookahead bias — structurally impossible rather than merely discouraged.

Architecture

Market data · signalsStrategiesPLUG-IN MODULESRisk gateNON-BYPASSABLEBrokerALPACA · PAPEREvent logAPPEND-ONLY · SYSTEM OF RECORD · NO UPDATE, NO DELETEProjectionsDERIVED STATEConsoleREACT · FASTAPIBackfillPOINT-IN-TIMETraining datasetsNO LOOKAHEAD BIAS

Results

Trading console dashboard showing an equity curve and five open paper positions
Operations console — demo environment, synthetic signals
Classified-signal feed with every demo row flagged SYNTHETIC by the provenance system
Signal feed — the provenance guard labels synthetic data and blocks it from real-broker decisions

Key Contributions

  • Append-only event log as the system of record; current state is derived as projections, never stored as mutable truth
  • Non-bypassable risk gate boundary in front of every order
  • Point-in-time historical backfill program with explicit lookahead-bias controls for building training datasets
  • Hexagonal ports-and-adapters architecture: adding a strategy or data source is one module plus one registration
  • React operations console over a Python / FastAPI / Postgres backend
FinanceMLSystemsPythonFastAPIPostgresReact