Built a full-lifecycle stat-arb platform solo — hexagonal architecture, 22-model ensemble, dual-broker execution. Here’s the full technical breakdown.

I've spent the last several months building Superintel — a personal quantitative trading platform built entirely solo. Here's what's under the hood:

**Architecture**

– Strict hexagonal (ports & adapters) architecture across 24 domain modules

– 31–32 FastAPI routers, ~145–150 endpoints

– Every layer is swap-swappable: broker, data source, model — without touching core logic

**ML Ensemble**

– 22-model prediction ensemble combining gradient boosting, LSTM, transformer-based models

– Features engineered from tick data, order book snapshots, and macro signals

– Ensemble voting with confidence thresholds before any signal is passed downstream

**Data Layer**

– TimescaleDB with 40 tables, 20 hypertables for time-series efficiency

– Real-time ingestion pipeline with deduplication and gap-fill logic

**Execution**

– Dual-broker execution with failover logic

– Human-in-the-loop approval gate before live order submission

– Risk gating layer checks position limits, drawdown, and volatility regime before execution

**Quality**

– 2,692 passing tests with a full DDD compliance suite

– Domain events, value objects, and aggregates enforced throughout

Happy to answer questions on architecture decisions, model selection, or how I structured the risk layer. What would you have done differently?