📈 FinSight Pro: Autonomous AI Financial Analyst

FinSight Pro is an end-to-end, multi-agent financial intelligence platform. It bridges the gap between raw quantitative data and accessible financial literacy by combining Machine Learning (XGBoost), Vector Retrieval (ChromaDB), and Generative AI (Llama-3 via Groq) into a seamless, cloud-deployed dashboard.

Unlike simple wrappers, FinSight acts as a fully autonomous research analyst. It triangulates technical price action, live market news, and statistical momentum to generate both professional-grade investment memos and "Explain Like I'm 5" (ELI5) translations for non-technical users.

---

🚀 Enterprise-Grade Features

• 🤖 LangGraph Multi-Agent Architecture:
  • Quant Node: Calculates real-time RSI, MACD, Moving Averages, and ATR.
  • RAG Node: Embeds and retrieves live company profiles and Yahoo Finance news using all-MiniLM-L6-v2.
  • Manager Node: Uses Llama-3-70b with strict regex-parsing to synthesize data into perfectly formatted Markdown reports.
• 🧠 On-Demand Machine Learning:
  • Features a cloud-executable ML pipeline that fetches 10 years of historical data, engineers features, and dynamically retrains an XGBoost Regressor to forecast the next closing price.
• 📊 Dual-Persona Insights: * Automatically generates a strict Wall Street Strategic Memo alongside a jargon-free "Beginner Translation" of the metrics.
• 📈 Vectorized Strategy Backtesting: * Simulates historical algorithmic trading (e.g., SMA Crossovers) vs. standard Buy & Hold strategies, outputting plain-English profit comparisons.
• ⚙️ CI/CD & MLOps: * Fully automated testing pipeline using Pytest and GitHub Actions to ensure feature engineering math remains stable before deployment.

---

🏗️ System Architecture

graph TD
    A[User Input: Ticker] --> B(Quant Engine)
    A --> C(Vector Ingestion)
    
    subgraph Data Engineering
    C -->|Fetch Live News| D[(ChromaDB)]
    end
    
    subgraph Machine Learning
    B -->|Calculate SMA, RSI, MACD| E[XGBoost Forecasting]
    B -->|Historical Prices| F[Vectorized Backtest]
    end
    
    subgraph Generative AI
    E --> G{LangGraph Orchestrator}
    D -->|Similarity Search| G
    G -->|Strict Prompting| H((Llama-3 70B))
    end
    
    H -->|Regex Parsed Output| I[Streamlit UI]
    F -->|Plotly Charts| I

Loading

---

🛠️ Tech Stack

• LLM & Orchestration: LangChain, LangGraph, Groq API (Llama-3).

• Machine Learning: XGBoost, Scikit-Learn, Pandas.

• Vector Database: ChromaDB, HuggingFace Sentence Transformers.

• Frontend UI: Streamlit, Custom HTML/CSS Metric Cards, Plotly.

• DevOps: GitHub Actions, Pytest.

---

⚡ Quick Start (Local Development)

1. Clone the Repository

git clone [https://github.com/pankrulez/finsight-ai.git](https://github.com/pankrulez/finsight-ai.git)
cd finsight-ai

2. Set Up Environment Variables Create a .env file in the root directory and add your Groq API key:

GROQ_API_KEY=gsk_your_free_key_here

3. Install Dependencies

pip install -r requirements.txt

4. Run the Application

Start the Streamlit server. The application will handle database ingestion and ML training dynamically via the UI.

streamlit run app.py

🧪 Automated Testing

This project utilizes continuous integration. To run the Pytest suite locally and verify the feature engineering math:

pytest tests/

---

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.