FinSight - Multi-Agent Financial Analysis Platform

FinSight is a sophisticated financial analysis platform that leverages multiple Large Language Models (LLMs) in a coordinated multi-agent architecture to provide real-time financial insights. The system processes natural language queries about publicly traded companies and delivers comprehensive analysis including investment recommendations, risk assessments, and financial metrics visualization.

🚀 Features

• Natural Language Processing: Ask questions about any publicly traded company in plain English
• Multi-Agent Architecture: Specialized AI agents for different financial analysis tasks
• Real-Time Data Integration: Live market data from Yahoo Finance and SEC EDGAR
• Interactive Visualizations: Rich charts for financial comparisons, risk scoring, and metric trends
• Dual-Mode Analysis: Beginner-friendly and professional-grade responses
• Document Processing: RAG-based analysis of financial documents (PDFs, 10-K, 10-Q filings)
• Time-Aware Queries: Intelligent parsing of time expressions like "past 5 years" or "Q3 2024"

🏗️ Architecture

FinSight uses a multi-agent system with five specialized agents:

1. Company Discovery Agent: Natural language entity extraction and company identification
2. Parser Agent: Financial data acquisition and structuring from APIs and documents
3. KPI Agent: Financial ratio and metric calculations (ROE, ROA, D/E, margins, etc.)
4. Risk Assessment Agent: Multi-factor risk analysis and scoring
5. Insight Generator Agent: Context-aware investment recommendations

System Architecture

┌─────────────────┐    ┌──────────────────────────────────────┐    ┌─────────────────┐
│   User Query    │───▶│        LangGraph Workflow            │───▶│    Response     │
└─────────────────┘    │                                      │    └─────────────────┘
                       │  ┌─────────────────────────────────┐ │
                       │  │      Company Discovery Agent    │ │
                       │  │      Parser Agent               │ │
                       │  │      KPI Agent                  │ │
                       │  │      Risk Assessment Agent      │ │
                       │  │      Insight Generator Agent    │ │
                       │  └─────────────────────────────────┘ │
                       └──────────────────────────────────────┘

Detailed System Architecture

🛠️ Technology Stack

Frontend

• Streamlit 1.28+ - Interactive web interface
• Plotly - Interactive visualizations and charts
• Session State Management - User history and API caching

Backend

• LangGraph - Multi-agent workflow orchestration
• LangChain - LLM integration and standardization
• Python 3.8+ - Core application logic

LLM Providers

• OpenAI GPT-4 - Insight generation and complex analysis
• Google Gemini 1.5 Flash - Company discovery and entity extraction
• Groq (Mixtral, Llama3-70B) - High-speed data processing and calculations

Data Sources

• Yahoo Finance API - Real-time market data and company fundamentals
• SEC EDGAR API - Official filings and XBRL financial statements
• OpenAI Embeddings + FAISS - Document processing and semantic search

📊 LLM Performance Results

Based on comprehensive evaluation across 5 financial analysis tasks:

Model	Overall Score	Avg Response Time	Best Use Cases
Gemini 1.5 Flash	0.70	2.91s	Company discovery, Document parsing
Mixtral (Groq)	0.68	1.60s	Risk assessment, Real-time responses
Llama 3 70B (Groq)	0.64	1.45s	KPI extraction, Basic parsing
GPT-4	0.53	7.25s	Insight generation, Complex analysis

🚦 Getting Started

Prerequisites

• Python 3.8 or higher
• API keys for:
  • OpenAI
  • Google (Gemini)
  • Groq

Installation

1. Clone the repository:

git clone https://github.com/your-username/finsight.git
cd finsight

2. Install dependencies:

pip install -r requirements.txt

3. Set up environment variables:

cp .env.example .env
# Edit .env with your API keys

4. Run the application:

streamlit run app.py

Environment Variables

Create a .env file with the following variables:

OPENAI_API_KEY=your_openai_api_key
GOOGLE_API_KEY=your_google_api_key
GROQ_API_KEY=your_groq_api_key

📖 Usage Examples

Basic Queries

• "What is Apple's current P/E ratio?"
• "Analyze Microsoft's financial performance over the past 5 years"
• "Should I invest in Tesla? Give me a risk assessment"

Advanced Queries

• "Compare Amazon's debt-to-equity ratio with the industry average"
• "What are the key risks for investing in Netflix right now?"
• "Analyze Google's quarterly earnings trends and provide investment insights"

Document Analysis

• Upload a company's 10-K filing and ask: "What are the main risk factors mentioned in this document?"
• "Summarize the key financial highlights from this earnings report"

🔧 Configuration

Model Selection Criteria

Each agent is optimized for specific tasks based on:

• Performance & Latency: Sub-second response times for discovery, <5s for analysis
• Task-Specific Capabilities: Specialized model selection for each agent type
• Cost Efficiency: 90%+ cost savings compared to using GPT-4 for all tasks
• Output Quality: Balanced approach between speed and accuracy

Acceptable Latency Thresholds

• Company Discovery: <2 seconds
• Analysis Agents: <5 seconds
• Full Analysis Pipeline: <10 seconds

🎯 Key Features

Time-Aware Query Processing

• Interprets natural language time expressions
• Handles fiscal vs. calendar year differences
• Accounts for typical 1-2 quarter reporting delays
• Intelligent fallback to most recent available data

Document Processing (RAG)

• Supports PDF documents (10-K, 10-Q, earnings reports)
• 1000-character chunks with 200-character overlap
• Semantic search returns top 5 most relevant chunks
• Focus on relevant sections rather than entire documents

Risk Assessment

• Quantitative risk scoring (Debt, Liquidity, Market risks)
• Qualitative factor analysis
• Multi-factor risk evaluation
• Risk level categorization and recommendations

🛡️ Error Handling

The system includes comprehensive error handling:

• Fallback responses for API failures
• User-friendly error messages
• Automatic retry mechanisms
• Graceful degradation when data is unavailable

📈 Performance Metrics

• Company Identification Accuracy: 80%+
• Data Extraction Success Rate: 95%+
• Average Cost per Query: $0.001-0.005 for discovery, $0.01-0.02 for full analysis
• Response Time: Sub-5 second end-to-end processing

🔮 Future Enhancements

Planned Features

• Integration with Bloomberg and Reuters for premium data
• Redis implementation for distributed caching
• Technical analysis and sentiment tracking agents
• Enhanced agent collaboration using CrewAI
• Parallel agent execution for improved performance
• User feedback loops for output refinement

Architecture Improvements

• Migration to CrewAI for enhanced agent collaboration
• Distributed caching with Redis
• Microservices architecture for better scalability
• Enhanced monitoring and logging capabilities

🙏 Acknowledgments

• OpenAI for GPT-4 and embedding models
• Google for Gemini 1.5 Flash
• Groq for high-speed inference infrastructure
• Yahoo Finance and SEC EDGAR for financial data APIs
• The open-source community for the underlying frameworks

Disclaimer: FinSight is designed for educational and research purposes. All financial analysis and investment recommendations should be verified with qualified financial professionals before making investment decisions.