AI-Powered Polymarket Trading Bot

An AI trading system that monitors Twitter for market-relevant content, processes it through an AI pipeline, and executes automated trades on Polymarket based on LLM analysis. I chose polymarket over the stock market due to a lack of traditional players, and subsequently higher inefficiencies.

Real-time dashboard showing live tweet monitoring and trading activity

What This Project Does

This system combines social media monitoring, AI analysis, and automated trading:

1. Twitter Monitoring: Scrapes Twitter content from specified accounts using browser automation
2. AI Content Analysis: Processes tweets through a LangGraph pipeline with web search and market relevance analysis
3. Vector Search Matching: Uses ChromaDB to find relevant prediction markets based on tweet content similarity
4. Automated Trading: Makes trading decisions using LLaMA 3.2 and executes trades on Polymarket
5. Real-time Price Tracking: Maintains current bid/ask prices via WebSocket connections

System Architecture

Data Flow

Twitter → driver.py → FastAPI Webhook → ChromaDB Storage ↓ ChromaDB → LangGraph Pipeline → Market Analysis → Trade Execution ↓ ↓ Web Search Enhancement PostgreSQL Logging

Core Components

1. driver.py: Twitter scraper using Chrome automation
2. twitterWebhook.py: FastAPI server handling tweet ingestion
3. langgraphPipe.py: AI analysis pipeline with structured outputs
4. websocketPoly.py: Real-time market price monitoring

AI Pipeline Steps

1. Content Enrichment: Web search via Tavily API
2. Vector Similarity: ChromaDB semantic matching
3. Market Analysis: LLM-based relevance scoring
4. Trade Decision: Structured output with reasoning
5. Significance Check: Evaluates potential for >5% price movement
6. Execution Logging: PostgreSQL trade records

Tech Stack

AI & Machine Learning

• LangGraph: Multi-step AI workflow orchestration
• LangChain: LLM framework and integrations
• Ollama: Local LLM inference (LLaMA 3.2)
• HuggingFace Transformers: Text embeddings (all-MiniLM-L6-v2)
• ChromaDB: Vector database for semantic search

Backend & APIs

• FastAPI: High-performance webhook server
• PostgreSQL: Structured data storage (markets, trades)
• Tavily: Web search enrichment API
• Polymarket API: Trading execution and market data

Automation & Frontend

• nodriver: Chrome browser automation for Twitter scraping
• WebSockets: Real-time Polymarket price feeds
• Next.js: React-based dashboard framework
• TypeScript: Type-safe frontend development
• Docker Compose: Multi-service orchestration

Frontend Dashboard

Real-time Next.js dashboard with comprehensive monitoring capabilities:

• Live Tweet Monitoring: Real-time display of incoming tweets
• Trading Activity Tracking: Visualization of executed and skipped trades
• Connection Status: Live backend connectivity indicator
• Real-time Updates: Server-Sent Events (SSE) for instant data streaming

Access Points:

• Dashboard: http://localhost:3000
• API: http://localhost:8000/api/recent, http://localhost:8000/events

Vector Embedding Visualization

• https://www.youtube.com/watch?v=6-y4AuxsqeA

How to Run

Environment Setup

1. Clone and navigate to the project:

bash

git clone <repository-url> cd polyAi/src/backend/expStuff

1. Create environment file:

bash

cp .env.example .env

1. Configure your .env file:

env

```

Database Configuration

POSTGRESUSER=yourpostgresuser POSTGRESPASSWORD=yoursecurepassword POSTGRESHOST=localhost POSTGRESPORT=5431 POSTGRESDB=polymarkettrading

API Keys

TAVILYAPIKEY=yourtavilyapikey POLYKEY=yourpolymarketapikey POLYADDRESS=yourpolymarketwallet_address

WebSocket Authentication

APIKEY=yourwebsocketapikey APISECRET=yourwebsocketsecret APIPASSPHRASE=yourwebsocketpassphrase ```

Start the Application

bash

docker-compose up --build

This will automatically start all services and launch the dashboard at http://localhost:3000.

Service Management

bash

```

View service status

docker-compose ps

Check logs

docker-compose logs -f [service-name]

Stop services

docker-compose down ```

Configuration Options

Monitored Twitter Users

Edit MONITORED_USERS in driver.py:

python

MONITORED_USERS = ["username1", "username2", "username3"]

Polling Frequency

Adjust POLL_INTERVAL in driver.py:

python

POLL_INTERVAL = 10 # seconds between Twitter checks

Model Configuration

Modify AI model in langgraphPipe.py:

python

llm = ChatOllama( model="llama3.2:latest", temperature=0, base_url="http://ollama:11434" )

Database Operations

Connect to PostgreSQL

bash

psql -h localhost -p 5431 -U ${POSTGRES_USER} -d ${POSTGRES_DB}

Database Schema

• markets: Prediction market metadata
• tokens: Market outcome tokens with pricing data
• bought: Trade execution log with tweet references

Monitoring & Debugging

API Endpoints

• Health Check: GET http://localhost:8000/db
• Tweet IDs: GET http://localhost:8000/tweet-ids
• Market Data: POST http://localhost:8000/poly

Data Persistence

• PostgreSQL: ./backend_pgdata/ volume
• ChromaDB: ./chroma/ directory
• Ollama Models: ollama_data volume
• HuggingFace Cache: huggingface_cache volume

Development Challenges

Browser Automation & Anti-Bot Detection

Twitter's anti-bot measures required implementing advanced stealth