GoBandit 🎯 - High-Performance Multi-Armed Bandit Engine in Go

GoBandit is a pet project, an experimentation engine that implements Thompson Sampling for intelligent A/B testing at scale. Built with Go, with a frontend written using HTMX, Templ and Tailwind, and PostgreSQL storage. It provides a clean, performant API for managing multiple concurrent experiments with sophisticated exploration/exploitation strategies.

Local Development

Start PostgreSQL:

docker compose up -d postgres

Start server:

go run main.go

Go to http://localhost:8080

API

POST /tests
Create a new A/B test with multiple arms

GET /tests/{testID}/arm
Get next arm using Thompson Sampling

POST /tests/{testID}/arms/{armID}/result
Record result for an arm pull

GET /tests/{testID}
Get test statistics and configuration

Performance Characteristics

• Time Complexity: O(n) for arm selection where n is number of arms (typically small)
• Space Complexity: O(1) per request
• Database:
  • Single query for arm selection
  • Single-row updates for result recording
  • Proper indexing for O(1) lookups

Usage Example

curl -X POST \
  http://localhost:8080/tests \
  -H 'Content-Type: application/json' \
  -d '{
    "name": "Random Test Name",
    "description": "This is a randomly generated test",
    "arms": [
      {
        "name": "Arm A",
        "description": "This is the first arm"
      },
      {
        "name": "Arm B",
        "description": "This is the second arm"
      }
    ]
  }'

Progress

• TODO: Architecture (maybe too simple to diagram?)
• TODO: Blogpost
• TODO: documentation
• TODO: Performance benchmarks
• TODO: Example use cases

Future Enhancements

• Real-time dashboard?
• Redis caching layer for high-traffic scenarios
• Additional bandit algorithms (UCB1, ε-greedy)
• Prometheus metrics