Sentinel is a real-time anomaly detection system that bridges the gap between Quantitative Finance (Numbers) and Qualitative Analysis (News). It monitors high-frequency market data for statistical anomalies (Flash Crashes) and, upon detection, instantly triggers an AI Agent to read simultaneous news headlines and explain the root cause of the crash.

• Real-Time Volatility Monitoring: Calculates rolling standard deviation using Pathway's temporal windowing.
• AI-Powered Analysis: Uses Google Gemini 2.0 Flash Lite to explain market anomalies by correlating them with breaking news.
• Live Dashboard: Interactive Streamlit dashboard for real-time visualization.
• Production-Ready Architecture: Built with Pathway, a high-performance stream processing framework.

The easiest way to run Sentinel is using Docker Compose:

1. Prerequisites: Install Docker Desktop.
2. Configuration: Ensure your GOOGLE_API_KEY is set in the .env file.
3. Run:

   docker-compose up --build

4. Access:
   • Dashboard: http://localhost:8501
   • Engine: Runs in the background.

If you prefer running manually on Windows/WSL:

1. Clone the Repository:

   git clone https://github.com/SanyamBK/Sentinel-Financial-Risk-Engine.git
   cd Sentinel-Financial-Risk-Engine

2. Install WSL (Windows Only):

   wsl --install -d Ubuntu

3. Install Dependencies in WSL:

   wsl -d Ubuntu python3 -m pip install --break-system-packages pathway google-generativeai python-dotenv

4. Install Streamlit (Windows):

   python -m venv .venv
   .venv\Scripts\activate
   pip install streamlit pandas plotly watchdog

5. Configure API Key: Create a .env file in the root directory:

   GOOGLE_API_KEY=your_gemini_api_key_here

6. Run the Demo:

   .\run_demo.ps1

While we leveraged the Pathway LLM App repository for architectural inspiration, we implemented a Custom User Defined Function (UDF) for the Agent integration rather than using a pre-built RAG template.

Reasoning:

1. Event-Driven Efficiency: Standard RAG templates are query-driven. Sentinel is event-driven. By using a custom UDF, we injected conditional logic (if volatility > 0.5) to trigger the LLM only during critical market events.
2. Latency Optimization: This approach minimizes API calls, reducing latency and ensuring the dashboard updates in sub-seconds without waiting for LLM inference on stable data rows.
3. Compliance: This implementation fully utilizes Pathway's xPack capabilities for reasoning over live streams while adhering to strict rate-limit constraints.

The system implements the following Pathway features as specified in the project requirements:

• Windowing: pw.windowby() with 30-second tumbling windows for volatility calculation.
• ASOF Join: join_interval() to attach latest news to price windows.
• UDFs: @pw.udf decorator for Gemini AI integration.
• Stream Replay: pw.demo.replay_csv() for simulating high-frequency data.