We built institutional-style options analytics for retail traders on FastAPI + React
I'm John, founder of GammVault.
The thing that pushed us to build it: the analytics hedge funds treat as table stakes dealer gamma exposure maps, options flow context, calibrated signal models are either buried inside five figure terminals or repackaged as hype in $99/mo Discord groups. There wasn't an honest middle. So we built one.
GammVault maps gamma exposure (GEX), scans options flow, and generates directional options signals buy call, buy put, sell call, sell put from a machine learning model. The part we care most about is calibration: the model is Platt calibrated so the confidence score actually reflects edge, instead of spitting out a meaningless 90% on everything. Signals only fire inside a disciplined confidence band. Most "AI trading" products optimize for how confident they sound. We optimized for whether the number is honest.
The stack, since this is IH:
Backend: FastAPI / Python 3.11, PostgreSQL, Redis, on Railway
ML: XGBoost, 30 features, volatility-regime + direction, probability-calibrated
Frontend: React 18 + Vite, mobile via Capacitor (Android live, iOS next)
Data: Polygon (15 minute delayed), Stripe for billing
You can also connect a supported broker (Alpaca, tastytrade) for AI-assisted execution the assistant stages a trade from a signal and helps you place it, with you in control the whole way.
A few honest constraints we hold ourselves to: the data is delayed, not real-time, and we say so everywhere. We don't make dark pool claims. And we won't ship a signal type until its backtested edge holds up we've got a futures model sitting in a drawer right now precisely because it doesn't.
Free tier to try it, Pro at $24.99/mo, Edge at $59.99/mo.
Bootstrapped, run under 0N3 BL0K INC. in Ontario. Happy to answer anything about the ML calibration, the Capacitor build pain, or the regulatory gray zone around broker connected execution that last one's been the hardest part by far.
