AlphaStream: Real-Time AI Sentiment Trading Pipeline

AlphaStream is a distributed trading infrastructure that bridges high-level AI inference with low-latency execution. It ingests live financial news, performs sentiment analysis using a transformer model, and routes trade signals to a high-performance C++ matching engine.

🏗️ System Architecture

• Inference Service (Python): Subscribes to Alpaca's real-time news WebSocket. It utilizes FinBERT (Hugging Face) to generate sentiment scores (-1.0 to 1.0) and translates them into trade actions (BUY/SELL/NEUTRAL).
• Execution Engine (C++): A high-performance service hosting a Limit Order Book (LOB). It listens for signals via ZeroMQ and executes trades using a price-time priority matching algorithm.
• Communication Layer (ZeroMQ): Uses a PUB/SUB pattern to decouple the high-latency AI layer from the low-latency execution layer.

🚀 Technical Highlights & Challenges

• Latency Optimization: Implemented ZeroMQ to ensure asynchronous communication, preventing the AI inference overhead from blocking matching engine operations.
• Cross-Platform Compatibility: Resolved GLIBC ABI versioning conflicts and library linking issues (nlohmann_json, libzmq) by engineering a multi-stage Docker build process with unified Debian bases.
• Defensive Engineering: Developed robust JSON parsing in C++ that utilizes pattern matching (find('{')) and key validation to ensure pipeline stability against malformed network packets.
• ML Performance: Leveraged Hugging Face Accelerate and MPS (Metal Performance Shaders) to optimize transformer inference on Apple Silicon hardware.

🛠️ How to Run

1. Clone the repository.
2. Add your ALPACA_API_KEY and ALPACA_SECRET_KEY to python_inference_service/.env.
3. Run the orchestrated environment:

   docker compose up --build