Summary

This project offers a production-grade MLOps pipeline for automated weekly stock report generation. It targets engineers and researchers needing automated financial analysis, leveraging Transfer Learning with LSTMs and Agentic AI (LangGraph) for accurate predictions and Bloomberg-quality reports. The system provides a modular, scalable, and observable architecture, shifting from traditional ML scripts to a robust, automated workflow.

How It Works

The core approach employs a Parent-Child transfer learning architecture, training a base LSTM model on the S&P 500 and fine-tuning it for individual stock predictions. A multi-agent system, orchestrated via LangGraph, simulates a financial analyst, market expert, and editor to synthesize diverse data sources (LSTM forecasts, news, sentiment) into comprehensive reports. This modular design enhances prediction accuracy and automates the complex process of financial report generation.

Quick Start & Requirements

• Primary install/run command: Clone the repository, sync dependencies with uv sync, then execute ./run_docker.sh or docker-compose up --build -d.
• Non-default prerequisites: Docker & Docker Compose, Ollama (running on host machine with gpt-oss:20b-cloud and nomic-embed-text models pulled), FinnHub API key, UV package manager.
• Estimated setup time or resource footprint: Setup involves Docker Compose orchestration of multiple services (FastAPI, Streamlit, Grafana, Ollama, etc.), suggesting a moderate resource footprint. Specific time is not estimated.
• Relevant pages: Streamlit UI: http://localhost:8501, Monitoring App: http://localhost:8502, Grafana: http://localhost:3000.

Highlighted Details

• Agentic AI Workflow: Features four specialized agents (Performance Analyst, Market Expert, Report Generator, Critic) coordinated by LangGraph for sophisticated report creation.
• Semantic Caching: Utilizes Qdrant vector database for embedding and caching reports, enabling instant retrieval for similar queries within a 24-hour window.
• MLOps Practices: Implements auto-healing API, model registry via DagsHub MLflow, and drift detection with Evidently AI for robust pipeline management.
• Transfer Learning: Leverages a pre-trained S&P 500 LSTM model (Parent) fine-tuned on specific tickers (Child) for efficient, accurate predictions.

Maintenance & Community

The project is marked as "Created with ❤️ by Karan." No specific community channels (e.g., Discord, Slack) or detailed contributor information beyond the primary author are provided in the README.

Licensing & Compatibility

Distributed under the MIT License. This license generally permits commercial use and integration into closed-source projects, with minimal restrictions beyond attribution.

Limitations & Caveats

The setup requires specific external services like Docker and Ollama to be running, along with necessary API keys. The README does not detail potential performance bottlenecks, known bugs, or unsupported platforms. The effectiveness of the transfer learning approach is dependent on the quality and relevance of the S&P 500 base model.