LangGraph Reflection Demo

This project demonstrates a simple LangGraph agent that:

1. Generates a short answer (5–10 sentences) to a user‑supplied question.
2. Critiques the answer for completeness, concreteness, and fluff.
3. If the critique indicates needs_revision, rewrites the answer up to a maximum number of rounds.

Features

• Separate nodes for drafting, reflecting, and rewriting.
• LLM‑based critic that returns a verdict (ok or needs_revision) and 2–3 critique points.
• Controlled loop: rewrites only if the verdict is needs_revision and the round count is below max_rounds.
• CLI interface: pass a question via -q or input interactively.
• Configurable maximum rounds via -m (default 2).

Requirements

• Python 3.10+
• langgraph
• langchain-openai

Install dependencies:

pip install -r requirements.txt

Usage

1. Set your OpenAI API key:

export OPENAI_API_KEY="your_api_key_here"

2. Run the demo:

python src/main.py -q "Explain the difference between a tool and a resource in MCP."

Or simply:

python src/main.py

and enter the question when prompted.

The script will output the final answer, the number of rounds performed, the verdict, and the critique points.

Project Structure

src/
├── main.py      # CLI entry point
├── graph.py     # LangGraph definition
└── nodes.py     # Node implementations
requirements.txt
README.md

License

MIT License