Repository navigation
🛠️ Release: v1.0.0 – Multi-Agent Question Routing API
Release Date: July 11, 2025
Author: @bmdarklight
Docker Image: bmdarklight/multi-agent-routing:latest
🚀 Overview
This is the first stable release of the LangGraph-powered multi-agent routing API, built with FastAPI and Docker. It uses an LLM-powered classifier to route user queries to specialized agents that solve math problems, answer code questions, or provide travel guidance.
✨ Features
✅ Core API
• POST /ask: Accepts a question and routes it to the appropriate agent.
• GET /health: Simple health check endpoint.
• GET /: Welcome message for the API.
🧠 Agent Orchestration (LangGraph)
• classifier_node: Uses an LLM to classify the question topic (math, code, travel, or unknown).
• Each agent node (math_bot, code_bot, travel_bot) is conditionally invoked via LangGraph based on the classifier output.
🤖 Agents
• MathBot: Solves math questions using LangChain’s LLMMathChain, with logic for interpreting fuzzy questions.
• CodeBot: Answers coding-related questions using LLM + search (mocked in tests).
• TravelBot: Recommends travel destinations based on user input.
🔁 CI/CD Pipeline
• Full GitHub Actions CI workflow:
• Installs dependencies and runs pytest
• Builds Docker image
• Publishes Docker image to Docker Hub on push to main
🧪 Test Coverage
• Unit and integration tests for:
• classifier.py
• Each agent (math_bot, code_bot, travel_bot)
• API routes using TestClient
• LLM-dependent logic is mocked to ensure fast, deterministic tests.
🐳 Docker
• Dockerfile supports building a full FastAPI app container.
• docker-compose.yml allows easy local setup.
• Image available on Docker Hub as:
bmdarklight/multi-agent-routing:latest
🛠️ How to Run Locally
docker build -t multi-agent-routing .
docker run -p 8000:8000 multi-agent-routing
Then open: http://localhost:8000/
📚 Future Plans
• Add support for more agents (e.g. WeatherBot, FinanceBot)
• Enable streaming responses from agents
• Enhance the UI for interacting with the API
• Integrate authentication (optional per deployment)