🧠 AI Financial Copilot

A production-grade, agentic AI system for financial decision-making. Featuring a multi-step LangGraph reasoning loop, real-time portfolio math, RAG-grounded expertise, and high-performance caching.

---

🌟 Key Features

🤖 Agentic Intelligence (LangGraph)

• Multi-Step Reasoning: Uses a cyclic state machine to decompose complex goals into discrete tasks.
• Tool Orchestration: Dynamically selects between Price data, News sentiment, Risk calculations, and RAG search.
• Parallel Dispatch: Executes independent tools (e.g., News and Market Data) simultaneously for ~50% lower latency.
• XML Thought Process: Transparent <thought_process> blocks show exactly how the agent is reasoning.

📐 Precision Financial Engine (No Hallucinations)

• Metric Verification: Real computation of Sharpe, Sortino, VaR (95%), Beta, and Max Drawdown using raw data.
• Explainable Methodology: Transparent metadata on every risk report showing time windows, benchmarks, and confidence intervals.
• Modern Portfolio Theory (MPT): Scipy-powered optimization for Max Sharpe, Min Volatility, and Risk Parity.

📚 RAG & Semantic Memory

• Knowledge Base: Indexed educational documents with ChromaDB to interpret why a metric is good or bad.
• Source Transparency: Dedicated source citation UI showing news articles and document references.
• Daily TTL Cache: High-performance lru_cache for stock data to avoid redundant API hits within the same trading day.

📰 Sentiment NLP (FinBERT)

• Financial NLP: Local ProsusAI/finbert model for accurate sentiment scoring of market news.
• Weighted Consensus: Aggregates multi-article news confidence for a more grounded "Verdict" on market direction.

---

🚀 Quick Start

1. Configure Environment

# Clone the repository
git clone https://github.com/your-username/AI_Financial_Copilot.git
cd AI_Financial_Copilot

# Set your API keys in .env
# Required: GROQ_API_KEY (from https://console.groq.com)

2. Run via Docker Compose (Recommended)

docker compose up --build

• Dashboard: http://localhost:8501
• API Docs: http://localhost:8000/docs

3. Local Development (Manual)

# Install dependencies
python -m venv venv
source venv/bin/activate  # or venv\Scripts\activate on Windows
pip install -r requirements.txt

# Start Backend
uvicorn app.main:app --reload

# Start Frontend
streamlit run dashboard/streamlit_app.py

---

🏗️ Technical Architecture

Layer	Technology
LLM Engine	Groq (llama-3.3-70b-versatile)
Orchestration	LangGraph, LangChain
Backend API	FastAPI, Pydantic, SlowAPI
Frontend	Streamlit, Plotly (Interactive Charts)
Data/Math	yfinance, SciPy, NumPy, Pandas
Vector DB	ChromaDB (RAG)
NLP	Transformers (FinBERT), sentence-transformers
Observability	LangSmith, Structured Logging

---

📈 Example Commands

The agent is designed to handle complex, nested requests:

• "What is a good Sharpe ratio? Also, analyze my portfolio: 40% AAPL, 60% MSFT."
• "Fetch the last 5 news articles for Tesla and tell me the overall sentiment sentiment."
• "Based on Modern Portfolio Theory, optimize for Min Volatility between NVDA, AMD, and INTC."

---

🛡️ License & Disclaimer

Licensed under MIT.
Disclaimer: This is an AI research tool for educational purposes only. It does not provide certified financial advice. Always consult a professional before making investment decisions.