Honegumi RAG Assistant

An agentic workflow for Bayesian optimization code generation, built with LangGraph and powered by OpenAI models. Honegumi RAG Assistant codifies the end-to-end pipeline—parameter extraction, skeleton code generation, documentation retrieval, and code synthesis—into reusable nodes orchestrated as a LangGraph. LangSmith integration tracks and visualizes your graph executions. The result? Complete, ready-to-run Bayesian optimization code generated in seconds from natural language descriptions.

🚀 Why Honegumi RAG Assistant?

• Agentic LangGraph design lets you describe problems in natural language and get production-ready code
• Fast iteration: eliminate manual coding and focus on science
• Built on Honegumi: Leverages deterministic skeleton generation from Honegumi
• Intelligent RAG: Retrieves relevant Ax Platform documentation to enhance code generation
• LangSmith-backed for graph tracking, versioning, and observability
• Production-ready: versionable, testable, pip-installable

🛠️ Prerequisites

• Conda (Miniconda or Anaconda)
• Python 3.11+
• OpenAI API key
• LangSmith API key (optional)

📘 Google Colab Tutorial

To help you get started quickly, we've prepared an interactive Google Colab tutorial:

Google Colab Tutorial: Getting Started with Honegumi RAG Assistant

In this tutorial, you'll learn how to:

• Install Honegumi RAG Assistant and all necessary dependencies on Colab
• Set up API keys using Colab Secrets
• Build a vector store from Ax Platform documentation
• Describe your optimization problem and generate code
• View the generated code in your Google Drive

The tutorial runs entirely in Colab—no local setup required. All you need is access to your Google Drive and valid OpenAI/LangSmith API keys.

Installation via pip

1. Create & activate a conda environment

   conda create -n honegumi_rag python=3.11 -y
   conda activate honegumi_rag
   

2. Install via pip

   pip install honegumi-rag-assistant
   

3. Configure your API keys

   Honegumi RAG Assistant will automatically look for a file named .env in your current working directory (or any parent) and load any keys it finds.

   In the folder where you'll run the CLI (or in any ancestor), create a file called .env containing:

   OPENAI_API_KEY=sk-...
   LANGCHAIN_API_KEY=lsv2_...
   

4. Build vector store (one-time setup)

   For best results with documentation retrieval, build the vector store:

   # Download the build script
   wget https://raw.githubusercontent.com/hasan-sayeed/honegumi_rag_assistant/main/scripts/build_vector_store.py
   
   # Run it
   python build_vector_store.py --output ./ax_docs_vectorstore
   
   # Set the path in your .env
   echo "AX_DOCS_VECTORSTORE_PATH=./ax_docs_vectorstore" >> .env
   

5. Run the assistant

   Now, start the honegumi-rag pipeline:

   honegumi-rag
   

   Honegumi RAG Assistant will:

   • Prompt you to describe your Bayesian optimization problem
   • Extract parameters (objectives, constraints, search space, etc.)
   • Generate a deterministic code skeleton using Honegumi
   • Retrieve relevant Ax Platform documentation
   • Generate complete, runnable Python code
   • Stream the code generation in real-time
   • Optionally save the generated code to a file

Usage Examples

Interactive Mode (Default)

honegumi-rag

You'll be prompted:

Please describe your Bayesian optimization problem.
(Press Enter when finished)

Your problem:

The assistant generates complete code and displays it in real-time.

Save Generated Code

honegumi-rag --output-dir ./my_experiments

Use Different Models

Customize which GPT models are used for each agent:

honegumi-rag \
  --code-writer-model gpt-5 \
  --param-selector-model gpt-4o \
  --retrieval-planner-model gpt-4o

Key Features

Multi-Agent Architecture

• Parameter Selector: Extracts optimization parameters from natural language
• Skeleton Generator: Uses Honegumi for deterministic code templates
• Retrieval Planner: Generates intelligent documentation queries
• Parallel Retrievers: Concurrent documentation retrieval for speed
• Code Writer: GPT-5 powered code generation with streaming
• Reviewer (optional): Quality assessment and refinement

Advanced Features

• LangSmith Integration: Full tracing and debugging support
• Streaming Output: See code generation in real-time
• Flexible Models: Mix GPT-5 and GPT-4o for cost-performance optimization
• Optional Save: Print code or save to file—your choice

Command Line Arguments

Example Problems

Chemical Process Optimization

Optimize temperature (100-300°C), pressure (1-5 bar), and catalyst concentration (0.1-1.0 M) 
to maximize conversion rate in a catalytic reaction.

Materials Design

Optimize composition of a polymer blend: Component A (0-100%), Component B (0-100%), 
and curing temperature (80-150°C) to maximize tensile strength while minimizing cost.

Machine Learning Hyperparameters

Optimize neural network hyperparameters: learning rate (1e-5 to 1e-1), 
batch size (16 to 256), and dropout rate (0.1 to 0.5) to maximize validation accuracy.

Documentation & Support

• Full Documentation: GitHub Repository
• Google Colab Tutorial: Interactive Tutorial
• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Email: hasan.sayeed@utah.edu

Feedback & Feature Requests

This project demonstrates a proof of concept of what's possible with agentic systems for Bayesian optimization. While Honegumi RAG Assistant works out-of-the-box for many scenarios, your use case may involve more complex pipelines, custom constraints, or specific modeling needs.

Have something bigger in mind? Want Honegumi RAG Assistant to handle advanced features or integrate with your workflow?

We'd love to hear from you!

• Open a GitHub issue
• Start a discussion
• Or reach out directly at hasan.sayeed.71.93@gmail.com

License

MIT License - see LICENSE.txt for details.

Acknowledgments

• Built with PyScaffold
• Powered by LangGraph and LangChain
• Skeleton generation by Honegumi
• Uses Meta's Ax Platform for Bayesian optimization