Case Study
Quant Engine
Institutional-grade systematic trading platform. Signal factory generating 10,000+ candidate signals, GPU-accelerated backtesting, Bayesian self-learning loop, and a 9-multiplier position sizing chain. Running live on Alpaca.
Tech Stack
Problem
Institutional-grade systematic trading is locked behind hedge fund infrastructure — expensive, opaque, and inaccessible to solo operators. The edge is in signal quality, risk management, and self-learning, not in trading terminal access.
Approach
Built five pillars: a signal factory (10,000+ candidates, walk-forward validation, Benjamini-Hochberg filtering), GPU-accelerated backtesting (PyTorch, 100K parallel simulations), real-time streaming (WebSocket + Alpaca), a 9-multiplier position sizing chain (vol, regime, FOMC, GEX, correlation), and a Bayesian self-learning loop with Thompson Sampling. NexusWatch feeds geopolitical risk directly into the sizing model.
Outcome
Running live on Alpaca with a paper portfolio. 666 tests, 170+ modules, ~50K LOC. Path: 30-day Sharpe estimate → 6-month GO/NO-GO for real capital. NexusWatch API productization adds $500–2K/mo per subscriber as a parallel revenue stream.
Key Highlights
- Signal factory: 10,000+ candidate signals with Benjamini-Hochberg multiple testing correction
- GPU-accelerated backtesting via PyTorch — 100,000 parallel simulations
- Bayesian feedback loop with Thompson Sampling meta-learner for dynamic strategy allocation