Open Alpha — Quantitative Atelier

Enterprise-grade quantitative factor research terminal. Atelier-grade design, production-grade engineering.

🎨 Live showcase: genius-apple.github.io/open-alpha — interactive Atelier tour with animations.

---

What is this?

Open Alpha is a full-stack quantitative research platform — from market data ingestion to factor synthesis, backtesting, live risk control, and portfolio research. It pairs production-grade engineering (microservices, TimescaleDB, Redis Streams, Numba JIT factor compiler) with atelier-grade design (editorial typography, amber-on-charcoal palette, monospace data precision).

Built for quants who care about both signal quality and the experience of working with the tools.

---

Showcase

Market Cartography — multi-frequency OHLC browser with Atelier sidebar	Loaded chart — real OHLC candles rendered
Factor Synthesis — DSL editor + quantile backtest workspace	Layered returns — quantile equity, cumulative IC, factor radar
Research Suite — Bayesian IC decay, posterior P(positive), rolling Sharpe / MaxDD

📐 More screenshots (risk panel, loaded charts)

Risk Panel (graceful degradation)	Chart loaded

---

Atelier Design Language

The frontend follows a deliberate aesthetic direction — not generic AI dashboard.

Element	Choice	Why
Display font	Fraunces (serif)	Editorial gravitas, FT-headline feel
Data font	JetBrains Mono	Precision; tabular nums for aligned metrics
Body font	Manrope	Modern, neutral, highly legible
Palette	Charcoal #0a0a0c + Amber #d4a55c + Rust #c4513a	Financial classicism (gold/value) over cliché emerald-on-black
Backgrounds	Radial amber glow + noise grain + data grid	Atmosphere & depth, not flat slabs
Cards	atelier-card with 1px amber gradient top-line	Detail = taste

---

Architecture

┌─────────────────────────────────────────────────────────────┐
│                    Frontend (React + Vite)                  │
│   Data Explorer · Factor Lab · Research Suite · Risk Panel  │
└────────────────────────────┬────────────────────────────────┘
                             │ HTTP / WebSocket
┌────────────────────────────▼────────────────────────────────┐
│                   API Gateway (port 8000)                   │
└──┬──────────────┬───────────────┬───────────────┬───────────┘
   │              │               │               │
   ▼              ▼               ▼               ▼
┌──────┐    ┌─────────┐    ┌───────────┐   ┌───────────┐
│Signal│    │  Risk   │    │ Execution │   │  Research │
│:8001 │    │  :8002  │    │   :8003   │   │  (shared) │
└──┬───┘    └────┬────┘    └─────┬─────┘   └───────────┘
   │             │               │
   └──────┬──────┴───────┬──────┘
          │  Redis Streams│
          ▼              ▼
   ┌────────────┐  ┌──────────────┐
   │ TimescaleDB│  │    Redis     │
   │  (hypertable)│ │ (streams+cache)│
   └────────────┘  └──────────────┘

Backend services

• Gateway — request routing, aggregation, graceful degradation
• Signal — factor signal generation (Numba-compiled DSL)
• Risk — volatility targeting, drawdown circuit breaker, liquidity guard
• Execution — DES backtester, order book, PnL tracking

Shared libraries (services/shared/)

• factor/ — DSL parser + Numba JIT compiler (no eval())
• marketdata/ — CCXT/FIX/ITCH collectors, lookahead guard, freq alignment
• backtest/ — DES high-fidelity simulator, LOB, impact model
• risk/ — risk manager, drawdown breaker, audit log
• research/ — Bayesian IC decay, cointegration, portfolio optimization
• db/ — TimescaleStore (hypertable partitioning)
• messagebus/ — Redis Streams + consumer groups

---

Test Pyramid (87% coverage)

Layer	What it catches	Tests
L1 Unit	Algorithm correctness (LOB matching, impact nonlinearity)	42
L2 Benchmark	Performance regression (<1s factor compute)	3
L3 BDD Acceptance	Spec semantic violations (lookahead leak, eval removal)	10
L4 API Contract	Frontend/backend field drift	9
L5 Fault Injection	Resilience (single-service-down isolation)	8
L6 DB E2E	Hypertable partitioning, query pruning	7 (skip if no Docker)
L7 User Journey	Real user task completion	13

✅ 82 passed / 7 skipped / 0 failed

Run locally:

python scripts/run_test_pyramid.py            # full pyramid
python -m pytest tests/test_user_journeys.py  # real user flows (needs stack up)

---

Quick Start

Prerequisites

• Python 3.11+
• Node 20+
• Docker (optional, for TimescaleDB/Redis)

Backend

pip install -r requirements.txt
python -m uvicorn backend.main:app --port 8010     # legacy API
python -m uvicorn services.gateway.app:app --port 8000  # gateway

Frontend

cd frontend
npm install
npm run dev    # http://localhost:5173

Full stack (Docker)

docker compose up -d timescaledb redis
./start_platform.bat   # Windows

---

Key Engineering Decisions

• No eval() — Factor DSL uses AST + Numba JIT. __import__/os.system are statically rejected.
• Lookahead guard — LookaheadGuard filters future data; freq_align shifts low-freq before ffill to prevent leakage.
• Graceful degradation — Every service degrades (not crashes) when dependencies fail. Gateway returns 503 with clear detail.
• DES over vectorized — High-fidelity discrete-event simulator replaces naive vectorized backtest.

---

Documentation

• Spec — 16 scenarios, requirements
• Coverage report — 87% verified, honest gap analysis
• Design screenshots — Atelier UI evidence

---

License

MIT — see LICENSE.