Agentic RAG Microservice Example with LangGraph

This repository demonstrates how to build an agentic microservice for Retrieval-Augmented Generation (RAG) using LangGraph. The service combines autonomous agent reasoning with retrieval and generation, providing a robust pattern for building intelligent, self-improving APIs.

---

What is an Agentic RAG Microservice?

An agentic microservice is a self-contained service that embeds an intelligent agent capable of:

• Understanding and enhancing user queries
• Retrieving relevant context from a knowledge base or vector database
• Generating answers using a language model
• Assessing the quality or groundedness of its own answers
• Iteratively refining its process until a satisfactory, grounded answer is produced

This pattern goes beyond traditional microservices by enabling adaptive, multi-step reasoning and self-correction within a single API endpoint.

---

How This Example Works

This repo implements an agentic RAG microservice as a FastAPI application with a /trigger endpoint. The workflow is orchestrated by LangGraph and consists of the following loop:

┌─► enhance_query ─► retrieve ─► generate_answer ─► ground_check ─┐
|                                                                |
└────────── repeat while ground_check == "NOT GROUNDED" ─────────┘

Key components:

• Query Enhancement: Improves the user's question for better retrieval.
• Retriever: Finds relevant context from a vector database (custom DfEmbedder).
• Generator: Uses a local language model (Ollama qwen3:8b by default) to answer.
• Grounding Check: Verifies if the answer is well-supported by retrieved evidence.
• Self-Improvement Loop: If not grounded, the agent refines its process and tries again.

Directory structure:

• selfrag/config.py – Central configuration for models and retrievers
• selfrag/graph.py – LangGraph workflow definition
• nodes/ – Modular steps for each node in the agent loop

---

Quick Start

1. Install dependencies:

   pip install -r requirements.txt
   # (Or use poetry with pyproject.toml)

2. Set up Ollama (local LLM):

   export OLLAMA_HOST=http://localhost:11434
   # Adjust if your Ollama instance runs elsewhere

3. Run the CLI demo:

   python examples/demo_cli.py

---

Running as an API Microservice

Start the FastAPI server:

uvicorn api:app --reload

Trigger the agent via HTTP:

curl -X POST \
  -H "Content-Type: application/json" \
  -d '{"query": "Who directed Inception?"}' \
  http://localhost:8000/trigger

Sample response:

{
  "answer": "Christopher Nolan"
}

---

Customization

• Change the language model: Edit the model name in selfrag/config.py.
• Swap out the retriever: Replace or extend the DfEmbedder logic.
• Modify the workflow: Adjust the LangGraph definition in selfrag/graph.py or add new nodes in nodes/.

---

Why Use This Pattern?

• Autonomy: The agent can reason, self-correct, and adapt its answers.
• Modularity: Each step is a node, making it easy to extend or swap components.
• Production-ready: Exposes a simple API endpoint for integration into larger systems.

---

References

• LangGraph
• LangChain
• Ollama
• DfEmbedder (custom vector DB)

---

License

MIT