Stock Agent

A multi-agent stock research system. Type a company name, get a full technical analysis with 5 charts.

Walks through: Orchestrator → Data flow → Researcher → Analyst → Sandbox → Full lifecycle

Stages

Stage	Folder	Status
1 — LangChain multi-agent	1-deepagent/	✅ Done
2 — MCP server	2-mcp-server/	✅ Done · Open interactive explainer
3 — Agent-to-agent (A2A)	3-a2a/	✅ Done · Open interactive explainer

Part 1 of this project is complete. All three stages (LangChain multi-agent, MCP server, A2A protocol) are fully working.

Part 2 is underway — AWS-hosted, continuously evolving with new agents.

Architecture (3-a2a)

The system runs five independent A2A services coordinated by an orchestrator:

User query
    │
    ▼
Orchestrator (main.py)
    │
    ├── Researcher Agent  (port 8001) — fetches price, fundamentals, news via MCP
    │
    ├── Analyst Agent     (port 8002) — technical analysis + 5 charts
    │
    └── Sentiment Agent   (port 8003) — StockTwits sentiment (runs in parallel with analyst)

Sentiment Agent

subagents/sentiment_agent.py — A2A service on port 8003.

Fetches social media data from StockTwits and uses GPT-4o-mini to extract investment-relevant sentiment:

• Fetches 2 pages (up to 60 messages) per ticker — safe rate limit for bulk use (40 req/2 min, well within the ~200/hr free tier)
• Tags each message as Bullish, Bearish, or Neutral based on StockTwits metadata
• GPT-4o-mini extracts overall sentiment, confidence level, and 3–5 key themes from the messages
• Result includes counts and themes displayed in a dedicated panel on the dashboard
• Runs in parallel with the Analyst agent to keep total latency down

Version 2 — Roadmap

Phase 1 — AWS Hosting ✅

• Deployed on AWS EC2 (t3.small, ap-south-1)
• Docker + Docker Compose for container orchestration
• ECR for Docker image registry
• GitHub Actions CI/CD pipeline (auto deploy on push to main)
• AWS Parameter Store for secrets management
• Nginx reverse proxy

Phase 2 — StockTwits Sentiment Agent ✅

• sentiment_agent.py — A2A service on port 8003
• Fetches StockTwits messages per ticker (2 pages = 60 messages)
• GPT-4o-mini extracts themes and overall sentiment
• Sentiment panel displayed on dashboard
• Runs in parallel with researcher + analyst

Phase 3 — Value Investment Agent (Alpaca) 🔄

• Alpaca paper trading account integration
• trading_agent.py — A2A service
• Value screening: P/E, ROE, profit margins, free cash flow
• Position sizing rules (max 10% per stock, 20% cash reserve)
• Stop loss (-15%) and take profit (+30%) rules
• Paper trades on Alpaca

Phase 4 — Portfolio Summary + WhatsApp Notifications

• Connect to Alpaca API directly
• Daily + weekly portfolio performance summary
• P&L breakdown per position
• All buy/sell orders fetched from Alpaca (no separate DB needed)
• WhatsApp notifications via Twilio Sandbox (free, personal use)
  • EOD report sent to your WhatsApp automatically
  • Buy/sell execution alerts in real time
  • On demand portfolio summary

Phase 5 — Multimodal Chart Analysis

• Feed generated charts to GPT-4o vision
• Visual pattern observations (head & shoulders, support/resistance)
• Pattern analysis added to final report

Phase 6 — Trading Suggestion Agent

• Short term trade suggestions (2-4hr or 1-3 day timeline)
• Separate from value investment (Phase 3 is long term)
• Based on technical indicators + sentiment signals