README.md

metadata

title: Financial Research Agent
emoji: 📊
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 5.49.0
app_file: app.py
pinned: false
license: mit

📊 Financial Research Agent

Multi-agent equity analysis combining SEC filings with real-time market intelligence

🔗 Try Live Demo →

---

🎯 What This Does

A production-grade financial analysis system that combines:

• Domain-specific ML models (FinBERT, SEC-BERT) for fundamental analysis
• Gemini 2.0 Flash for intelligent synthesis and reasoning
• LIME explainability to show which words drive predictions
• Multi-agent orchestration for comprehensive equity research

Architecture Flow

graph LR
    A[User Input: Ticker] --> B[SEC Filing Agent]
    A --> C[Market Intelligence Agent]
    B --> D[FinBERT/SEC-BERT Sentiment]
    B --> E[LIME Explainability]
    C --> F[yfinance Market Data]
    C --> G[NewsAPI Articles]
    G --> H[Gemini News Sentiment]
    D --> I[Synthesis Agent]
    E --> I
    F --> I
    H --> I
    I --> J[Gemini 2.0 Flash]
    J --> K[Investment Recommendation]

Agent Pipeline:

1. SEC Filing Agent: Analyzes 10-K/10-Q/8-K filings with FinBERT + SEC-BERT, extracts actual risk sentences
2. Market Intelligence Agent: Gathers real-time price data, technicals (RSI/MACD), and news with Gemini-powered sentiment
3. Synthesis Agent: Gemini 2.0 Flash cross-references fundamentals vs. market action for final recommendation

🚀 How to Use

1. Enter a stock ticker (e.g., TSLA, AAPL, RBLX)
2. Optionally add company name for better news search
3. Select SEC filing type (10-K for annual, 10-Q for quarterly)
4. Click Analyze
5. Wait ~1-2 minutes for comprehensive analysis

🔬 What Makes This Different

Unlike basic sentiment tools, this platform:

• ✅ Hybrid Intelligence: Combines domain-specific ML (FinBERT/SEC-BERT) with LLM reasoning (Gemini 2.0 Flash)
• ✅ Explainable AI: LIME shows which words drive sentiment predictions
• ✅ Actual SEC Filings: Extracts real risk sentences from 10-K/10-Q/8-K, not generic summaries
• ✅ Cross-Validation: What companies say (filings) vs. what markets do (price/news)
• ✅ Deterministic: Reproducible results with fixed random seeds for consistent analysis

⚙️ Configuration

Required Environment Variables:

• SEC_EMAIL: Your email (SEC EDGAR compliance)
• NEWS_API_KEY: NewsAPI key (get free tier)
• GEMINI_API_KEY: Google Gemini API key (get free tier)

Note:

• First run downloads ML models (~500MB) - subsequent analyses are much faster
• Without Gemini API key, falls back to rule-based synthesis (less sophisticated)

🛠️ Tech Stack

AI/ML:

• Domain-Specific Models: FinBERT, SEC-BERT (financial sentiment analysis)
• LLM: Gemini 2.0 Flash (synthesis and reasoning)
• Explainability: LIME (Local Interpretable Model-agnostic Explanations)

Data Sources:

• SEC EDGAR (official filings)
• yfinance (market data)
• NewsAPI (news sentiment)

Infrastructure:

• Framework: Custom multi-agent orchestration
• Interface: Gradio 5.0+
• Deployment: Hugging Face Spaces
• Language: Python 3.10+

✨ Key Features

• 📄 Multi-Filing Support: Analyze 10-K (annual), 10-Q (quarterly), or 8-K (current) reports
• 🧠 Hybrid AI: Domain-specific models + LLM reasoning for best of both worlds
• 🔍 Explainable Results: See which words influenced sentiment predictions
• 📊 Comprehensive Analysis: Fundamental (SEC) + Technical (RSI/MACD) + News sentiment
• 🎯 Risk Extraction: Extracts actual risk sentences from filings, not generic summaries
• 🔄 Deterministic: Same ticker = same results (reproducible with fixed seeds)

📝 Limitations

• SEC filings updated quarterly/annually (not real-time)
• NewsAPI free tier: 100 requests/day
• First analysis takes longer (model download ~500MB)
• Analysis time: 1-3 minutes depending on data availability
• yfinance may occasionally face rate limiting

🚀 Local Development

# Clone repository
git clone https://github.com/SanchitSharma10/financial-research-agent.git
cd financial-research-agent

# Install dependencies
pip install -r requirements.txt

# Set environment variables
export SEC_EMAIL="[email protected]"
export NEWS_API_KEY="your_newsapi_key"
export GEMINI_API_KEY="your_gemini_key"  # Optional

# Run locally
python app.py

Visit http://localhost:7860 in your browser.

🔗 Links

• Live Demo: Hugging Face Spaces
• GitHub: SanchitSharma10/financial-research-agent
• Author: Sanchit Sharma

---

⚠️ Important Disclaimer

This tool is for research and educational purposes only. NOT financial advice.

• Do not make investment decisions based solely on this analysis
• Always conduct your own due diligence and consult with financial professionals
• Past performance does not guarantee future results
• The creators assume no liability for financial decisions made using this tool

📄 License

MIT License - See LICENSE file