Finance AI Agent 💹

Autonomous 4-agent investment analysis system. Bull, Bear, Risk Analyst, and CIO Decision Maker debate a stock — then produce a professional PDF investment report with human-in-loop approval.

---

What It Does

Enter a stock ticker. The system launches 4 AI analysts in parallel who debate the investment thesis, assess risk across 5 dimensions, and deliver a structured verdict — all without human intervention until the final approval gate.

---

Agent Architecture

                    Stock Ticker Input
                          │
              ┌───────────┴───────────┐
              ▼                       ▼
        ┌──────────┐           ┌──────────┐
        │  Bull    │           │  Bear    │
        │  Agent   │           │  Agent   │
        │          │           │          │
        │ Bullish  │           │ Bearish  │
        │ signals  │           │ signals  │
        └────┬─────┘           └─────┬────┘
             │                       │
             └──────────┬────────────┘
                        ▼
               ┌─────────────────┐
               │  Risk Analyst   │
               │     Agent       │
               │                 │
               │ 5-dim risk      │
               │ assessment      │
               └────────┬────────┘
                        │
                        ▼
               ┌─────────────────┐
               │  CIO Decision   │
               │   Maker Agent   │
               │                 │
               │ Final verdict + │
               │ confidence      │
               └────────┬────────┘
                        │
                        ▼
               Human Approval Gate
                        │
                        ▼
               PDF Investment Report

---

Features

• 4-Agent Debate — Bull and Bear agents argue from opposing positions using real market data
• Real-time Data — Yahoo Finance price/fundamentals + SEC EDGAR XBRL filings
• 5-Dimension Risk — Market, liquidity, regulatory, operational, macro risk scoring
• Human-in-Loop — CIO verdict staged for human approval before PDF generation
• Backtesting — evals/ module for historical verdict accuracy testing
• PDF Reports — Professional investment reports generated with ReportLab
• Streamlit UI — Clean web interface for stock input and report viewing
• Parallel Execution — Bull and Bear agents run concurrently via LangGraph

---

Tech Stack

Component	Technology
Agent Orchestration	LangGraph (StateGraph, parallel nodes)
Primary LLM	Groq — Llama 3.3 70B
Fallback LLM	OpenAI GPT-4o
Financial Data	yfinance, SEC EDGAR XBRL API
Web Search	Tavily Search API
UI	Streamlit
PDF Generation	ReportLab
Language	Python

---

Quick Start

git clone https://github.com/iampriyabrat14/finance-ai-agent.git
cd finance-ai-agent
pip install -r requirements.txt

# Set environment variables
export GROQ_API_KEY=your_groq_key
export OPENAI_API_KEY=your_openai_key      # optional fallback
export TAVILY_API_KEY=your_tavily_key

# Run the Streamlit app
streamlit run ui/app.py

---

Project Structure

finance-ai-agent/
├── agents/           # Bull, Bear, Risk, CIO agent implementations
├── graph/            # LangGraph workflow (StateGraph definition)
├── prompts/          # System prompts per agent role
├── tools/            # yfinance, SEC, Tavily, PDF generation tools
├── evals/            # Backtesting + verdict evaluation
├── ui/               # Streamlit frontend
└── requirements.txt

---

Example Output

The system produces a PDF report containing:

• Executive summary with investment verdict
• Bull case: top 5 bullish signals with data
• Bear case: top 5 risk factors with data
• 5-dimension risk scorecard
• Historical XBRL financial snapshots
• Final CIO recommendation with confidence score

---

License

MIT © 2026 Priyabrat Dalbehera

---

Connect

Priyabrat Dalbehera — AI Engineer | Building production GenAI systems