ai-trading-paper

AI Trading Research & Paper-Trading Platform

V0.5 — Intelligent Stock Screener

V0.5 builds on the V0.4 multi-symbol foundation and adds a configurable rule-based stock screener.

Added

• Custom AND-condition screener
• Price, RSI, relative-volume, return, volatility and SMA-distance filters
• Saved scan definitions in configs/saved_scans.json
• Latest-match results across all downloaded symbols
• Screener unit tests
• V0.4 watchlists, scanners and comparisons retained

Example scan

• Close > 0
• RSI (14) < 70
• Relative Volume > 1.5
• 20D Return (%) > 0
• SMA 50 Distance (%) > 0

All conditions must pass for a symbol to match.

Run

pip install -r requirements.txt
python -m src.data.downloader
streamlit run app/dashboard.py
pytest -q

Scope

This release is a research screener, not an execution system. It does not place trades, claim predictive accuracy, or imply that historical technical conditions guarantee future performance.

V0.6 — Machine Learning Research

Adds a research-only ML layer with:

• forward-return labels with configurable horizon/threshold
• strict chronological train/test split
• Logistic Regression baseline
• Random Forest
• optional XGBoost
• time-series-aware sigmoid probability calibration
• accuracy, precision, recall, ROC AUC, Brier score and log loss
• model-specific feature importance
• latest-row probability snapshot
• Streamlit ML Research tab
• tests for label alignment and chronological splitting

Important: model probabilities and historical test metrics are research diagnostics, not guarantees of future returns or profitability.

Run:

pip install -r requirements.txt
pytest -q
streamlit run app/dashboard.py

V0.7 — AI Intelligence Engine

Adds a transparent intelligence layer on top of V0.6:

• ensemble model signal from calibrated probabilities
• confidence score based on directional strength and model agreement
• broad market-regime classification
• aggregated feature-based signal explanation
• model-health diagnostics using out-of-sample metrics
• Streamlit AI Intelligence tab

Confidence is a diagnostic score, not a probability of profit or a trading guarantee.

V0.8 — Scientific Backtesting

Adds a research-only backtesting engine in src/backtesting/backtester.py:

• next-bar-open execution to avoid lookahead
• commission and slippage
• fractional position sizing
• stop-loss / take-profit controls
• equity curve and trade journal
• return, CAGR, volatility, Sharpe, Sortino, drawdown, win-rate and profit-factor metrics
• chronological walk-forward evaluation
• Monte Carlo bootstrap of completed trade returns

The dashboard exposes the feature under 📈 V0.8 Scientific Backtesting. Backtests are simulations for research/paper trading only and do not imply future profitability.

V0.9 — Advanced Paper Trading

Adds src/paper_trading/engine.py with:

• virtual cash and positions
• simulated market orders with commission
• realized/unrealized P&L
• order status and reasons
• persistent JSON-compatible portfolio snapshots
• chronological decision replay using next-bar-open fills
• trade/order journal for auditability

The dashboard exposes this under 💼 V0.9 Advanced Paper Trading. No broker connectivity or live execution is included.

V1.0 Production + Mobile API

The platform now includes a FastAPI REST service for authenticated paper trading and mobile clients. It uses SQLAlchemy persistence (SQLite by default; PostgreSQL is supported through DATABASE_URL with the appropriate PostgreSQL driver), bearer-token authentication, CORS configuration, order/position/trade/decision endpoints, and optional webhook notifications. No broker execution is implemented.

Run API locally

uvicorn src.api.app:app --reload --port 8000

OpenAPI docs are available at /docs. Copy .env.example to .env and set a strong JWT_SECRET before deployment. For PostgreSQL, set DATABASE_URL and install a compatible SQLAlchemy PostgreSQL driver.

Docker

docker compose up --build

V1.1 — Production Hardening

• Security headers and request IDs
• Database readiness endpoint
• Login rate limiting
• Idempotent order retries
• Audit event trail
• Paginated order/trade/decision APIs
• Alembic migration for production databases
• Secure production configuration checks
• Mobile API contract

V1.4 — Production Infrastructure

• PostgreSQL + Redis Docker services
• Shared Redis rate limiting when REDIS_URL is configured
• Redis-backed job queue with local development fallback
• API request/error/latency metrics
• Mobile API contract hardened for unreliable networks
• Production deployment checklist in deploy/V1.4_PRODUCTION.md

V1.5 — Event-driven paper trading

V1.5 adds a background paper-trading monitor. It refreshes open-position quotes, updates mark-to-market prices, detects stop-loss/take-profit triggers, and exposes engine status/start/stop/run-once API endpoints. It remains paper-trading-only; risk triggers are reported to the application layer and no broker order is ever submitted.

V1.6.1 — Reliability fixes

• Fixed RSI warm-up behavior for one-sided price series so valid RSI values are produced instead of all-NaN output.
• Kept extreme close-to-close moves as validation warnings without cascading the same outlier into an OHLC hard error.
• Added regression coverage for the RSI behavior.

V1.6 — Strategy Automation

• Persistent per-user paper strategies with symbol lists and risk controls.
• Configurable minimum signal confidence, position sizing, maximum exposure, open-position cap, and cooldown.
• Start/stop/run-once strategy engine.
• Automated momentum-driven paper entries/exits with existing order/audit infrastructure.
• Deterministic strategy backtest endpoint for supplied price series.
• No broker execution; all automation remains paper-only.

V1.8.0 — Market Terminal + AI Mobile

• Mobile Markets terminal with auto-refreshing market snapshots.
• Broad discovery universe for Indian/global indices, Indian stocks, commodities, FX and crypto.
• Arbitrary Yahoo Finance symbols can be opened for supported instruments.
• Multi-range OHLC candlestick charts: 1D, 5D, 1M, 6M, 1Y, 5Y.
• AI Intelligence screen exposing market regime, RSI, MACD, SMA context and existing ML research diagnostics.
• Paper portfolio remains separate from market-data research and no broker execution is added.
• The default Yahoo Finance adapter is not an exchange-certified tick feed; a licensed real-time provider is required for guaranteed tick-level coverage.