AI Investment Committee Platform

This is an institutional-style portfolio intelligence MVP inspired by BlackRock's investment workflows. It acts as an automated "Investment Committee", fetching live market data, identifying macroeconomic regimes, generating AI-powered investment recommendations via the Groq LLM API, and persisting all historical intelligence to MongoDB for change-tracking and comparative analysis.

---

🌟 Core Features

1. Live Data Aggregation: Fetches real-time market data, pricing, market cap, and calculates annualized volatility using yfinance.
2. Regime Detection: Analyzes current macroeconomic indicators (inflation, GDP, interest rates) to classify the current market environment (e.g. "Stagflation Risk", "Stable Growth", "High Inflation").
3. AI-Driven Committee Analysis: Feeds aggregated structured data into a Groq-powered LLM (llama-3.3-70b-versatile). The AI is prompted to act as a senior quantitative finance architect, outputting pure JSON recommendations (Buy/Hold/Sell) with a confidence score, bull case, bear case, and key risks.
4. Historical Memory & Context: Every analysis is automatically saved to MongoDB Atlas. When a stock is analyzed multiple times, the system retrieves the previous analysis, compares it to the new one, and outputs a historical_context payload showing exactly how the recommendation, confidence, and risks have shifted over time.
5. Change Tracking & Auditability: Stores chronological shifts in a dedicated recommendation_changes database collection, allowing institutional memory to be queried via the API.
6. Multi-Asset Portfolio Analysis: Analyzes portfolios to determine sector exposure and concentration risks, providing holistic portfolio-level recommendations.

---

🏗️ Architecture

The platform follows a clean service-oriented architecture:

• Market Agent: Interacts with Yahoo Finance for live pricing and volatility metrics.
• Macro Agent: Provides macroeconomic baseline data (inflation, interest rates, GDP growth).
• Regime Service: Evaluates macro conditions and outputs the market regime.
• Portfolio Agent: Aggregates multi-asset portfolios for sector exposure and concentration risks.
• Committee Agent: The LLM integration layer (Groq API) responsible for qualitative reasoning.
• Comparison Service: The historical intelligence engine that calculates the diffs in rating, confidence, and newly identified risks between runs.
• MongoDB Service: Persists all analyses, portfolio runs, and change tracking logs to MongoDB.

---

🚀 Setup Instructions

Prerequisites

• Python 3.9+
• A valid Groq API Key
• A MongoDB Atlas cluster

Installation

1. Create a Virtual Environment

   python -m venv venv
   source venv/bin/activate

2. Install Dependencies

   pip install -r backend/requirements.txt

3. Configure Environment Variables Create a .env file in the root directory and add your Groq API key and MongoDB URI:

   GROQ_API_KEY=your_groq_api_key_here
   MONGODB_URI=mongodb+srv://<user>:<password>@cluster.mongodb.net/

4. Run the Application Start the FastAPI development server:

   uvicorn backend.main:app --reload

   The API will be accessible at http://localhost:8000. You can access the interactive Swagger UI at http://localhost:8000/docs.

---

📡 API Documentation

1. Analysis Endpoints

POST /analyze

Generates a complete market, macro, and AI committee analysis for a single ticker. If the ticker was analyzed previously, it automatically generates a historical_context comparison. Payload:

{
  "ticker": "RELIANCE.NS"
}

POST /analyze-portfolio

Generates an aggregated analysis for a multi-asset portfolio, including sector exposure and concentration risk. Payload:

{
  "portfolio": [
    { "ticker": "TCS.NS", "weight": 50 },
    { "ticker": "INFY.NS", "weight": 50 }
  ]
}

2. Historical Intelligence Endpoints

Every time an analysis is generated, it is persisted to MongoDB. These endpoints retrieve that memory.

• GET /history/stocks: Retrieves the most recent stock analyses across all tickers.
• GET /history/stocks/{ticker}: Retrieves the entire chronological timeline of analyses for a specific asset.
• GET /history/stocks/{ticker}/latest: Retrieves only the absolute latest analysis for a specific asset.
• GET /history/stocks/{ticker}/changes: Retrieves a chronological log of all recommendation, confidence, and risk changes that occurred for a specific asset.
• GET /history/portfolios: Retrieves the history of all portfolio analyses.
• GET /history/portfolios/latest: Retrieves the absolute latest portfolio run.

---

🗄️ Database Collections

• stock_analyses: Stores the raw inputs, macro data, and the final LLM committee output for every individual stock run.
• portfolio_analyses: Stores the portfolio compositions, sector exposures, concentration risks, and portfolio-level LLM recommendations.
• recommendation_changes: Dedicated audit log storing only the calculated differences (e.g., changes from "Buy" to "Hold", drops in confidence) between subsequent analysis runs for the same ticker.