ChatDocs for Loop

ChatDocs for Loop is a local AI assistant that helps developers, contributors, and users navigate the open-source Loop diabetes management app documentation.
It combines local embeddings and Gemini 2.5 Flash to deliver accurate, grounded answers -> and tells you when something isn’t in the docs.

Why This Project

Loop is a powerful open-source app that helps people with diabetes automate insulin delivery.
However, its documentation and codebase can be complex to navigate — especially for newcomers.

ChatDocs was built to bridge that gap: a local, Gemini-powered assistant that helps developers and contributors explore the Loop documentation and repositories conversationally.

This project demonstrates:

• Practical application of LLMs (Gemini) for technical documentation understanding
• Implementation of a Retrieval-Augmented Generation (RAG) pipeline
• Thoughtful UX for developer productivity and knowledge exploration using Makefile

Demo

Link to Demo

Example Use Cases

• “Where does Loop define insulin delivery logic in code?”
• “How do I configure a new CGM according to the LoopDocs?”
• “What dependencies does the LoopKit framework have?”
• “Explain how Loop handles carb entries.”

---

Features

• Context-aware answers powered by Gemini + LangChain, combining human-readable docs and actual source code for deeper reasoning.
• Local and private — all embeddings, retrieval, and chat run on your machine (no external API calls beyond Gemini).
• Transparent retrieval — see which document each answer came from.
• Streamlit interface for developers — simple, fast, and designed for documentation exploration.

---

Tech Stack

Component	Purpose
Gemini API	LLM for reasoning, summarization, and answering queries
LangChain	Retrieval-Augmented Generation (RAG) pipeline orchestration
Chroma	Local vector database for storing and querying document embeddings
Hugging Face (Sentence Transformers)	Embedding generation using all-MiniLM-L6-v2
Streamlit	Interactive web UI for chatting with documentation
Python	Core logic, environment, and integration scripting

Architecture

         ┌───────────────────────┐
         │  LoopDocs Markdown    │
         └────────────┬──────────┘
                      │
               Load & Split Docs
                      │
           Embed with HuggingFace
                      │
              Store in Chroma DB
                      │
                 Retrieve Context
                      │
    ┌─────────────────┴─────────────────┐
    │      ChatGoogleGenerativeAI       │
    │         (Gemini 2.5 Flash)        │
    └─────────────────┬─────────────────┘
                      │
                 Streamlit Chat UI

Developer Commands

Command	Description
make check-gemini	Verify your Gemini API key and accessible models
make setup	Create virtual environment and install dependencies
make refresh	Pull latest LoopDocs and rebuild embeddings
make inspect	Inspect Chroma vector database
make chat	Start ChatDocs in CLI mode
make run-ui	Launch Streamlit web app

Setup

Step 1 - Clone the repo

git clone https://github.com/jeaend/ChatDocs.git
cd ChatDocs

Step 2 - Create the environment

make setup

Step 3 - Add your Gemini API key, see the example .env file for guidance

Create a .env file in the root:

GOOGLE_API_KEY=your_gemini_api_key

You can get a key at https://makersuite.google.com/app/apikey

Step 4 - Verify Gemini access

make check-gemini

Step 5 - Build the vector store

make refresh

Step 6 - Start chatting

CLI version

make chat

Streamlit UI

make run-ui

Then open the Streamlit URL (usually http://localhost:8501) to start chatting with the Loop docs.

---

Project Structure

ChatDocs/
│
├── app.py              # Streamlit interface
├── config.py           # Environment and API key setup
├── requirements.txt
│
├── docs/               # LoopDocs markdown files
├── repos/              # Cloned LoopKit repositories
├── data/               # Local embeddings / vector indexes
│
└── utils/
    ├── loader_docs.py  # Loads LoopDocs
    ├── loader_code.py  # Loads and parses LoopKit repo files
    ├── embedder.py     # Embedding + Chroma store logic
    └── chat.py         # Retrieval + chat chain

---

Future Enhancements

• Multi-source documentation support (adding source code)
• Deployment to Streamlit Cloud / Hugging Face Spaces
• Option to switch between Gemini, OpenAI, and Claude

---

Author

Jeanne Endres - Data Science & Software Engineer
LinkedIn | GitHub | Website