Intelligent Stock Analysis using multi-agent RAG System for SEC Filings
Intelligent Stock Analysis — Multi-Agent RAG System for SEC Filings
The Challenge
Equity research is painfully manual. Analysts spend hours digging through 200+ page 10-K filings, cross-referencing financial APIs, and stitching together technical indicators — just to produce a single stock report. What if an AI system could do all of that in minutes?
The Solution
I built a multi-agent platform where specialized AI agents collaborate to automate the entire equity research workflow — from pulling SEC filings to generating technical analysis to producing a polished PDF report, all triggered by a single stock ticker.
Key Achievements
End-to-end automation — ticker in, full research report out
4+ financial data APIs integrated into a single unified pipeline
Intelligent document retrieval from 10-K filings using vector embeddings
Automated technical analysis with MACD, RSI, Bollinger Bands, and valuation models
How It Works
The system orchestrates multiple specialized agents, each responsible for a different piece of the research puzzle:
SEC Filing Agent — Pulls 10-K and 10-Q filings from SEC EDGAR, chunks them intelligently, and stores embeddings in Pinecone for fast semantic retrieval
Financial Data Agent — Fetches real-time and historical price data, fundamentals, and key ratios from Alpha Vantage API
Technical Analysis Agent — Computes indicators (MACD, RSI, Bollinger Bands) and generates visual charts for trend analysis
News & Sentiment Agent — Scrapes recent news via SERPAPI and runs sentiment analysis to capture market mood around the stock
Research Synthesizer — Gemini AI aggregates all agent outputs, reasons across them, and generates a structured equity research report with buy/hold/sell context
Report Generator — Compiles everything into a clean, downloadable PDF with charts, tables, and key insights
Technologies Used
AI/LLM: Gemini AI, LangChain, RAG architecture
Vector DB: Pinecone for semantic search across SEC filings
Data APIs: Alpha Vantage, SEC EDGAR, SERPAPI
Backend: Python, FastMCP for agent orchestration
Analysis: MACD, RSI, Bollinger Bands, DCF valuation
Output: Automated PDF report generation
Results
Metric | Manual Research | With This System |
|---|---|---|
Time per Stock Report | 4-6 hours | Under 10 minutes |
Data Sources Cross-Referenced | 1-2 at a time | 4+ simultaneously |
SEC Filing Analysis | Skim and keyword search | Semantic retrieval across full filings |
Consistency | Varies by analyst | Standardized methodology every time |
Impact
The average equity analyst covers 10-15 stocks. The bottleneck isn't analysis — it's data gathering and synthesis. This system eliminates the grunt work entirely, letting analysts focus on judgment calls rather than copy-pasting numbers from EDGAR. It's the difference between reading every page of a 10-K and asking it a question.
"Built to solve the problem I watched analysts face firsthand — spending more time finding data than thinking about what it means."
See It In Action
View on GitHub
