Simple Support Agent with RAG Tooling for Documentation Support

Simple Support agent primarily using RAG with Vector DB consisting of a companies documention with a fallback to OpenAI GPT-3.5-Turbo for general question handling.

To enable Retrieval-Augmented Generation (RAG) we do the following:

• Convert Q&A documents into embeddings using OpenAI’s text-embedding-ada-002 model.
• Store them in a Chroma vector store.
• At runtime, retrieve the most semantically similar Q&A using vector similarity search.
• If a strong match is found, use it directly. Otherwise, fall back to querying the OpenAI API.

This approach enables flexible, robust answers even for paraphrased questions.

---

Project Overview

This project allows you to:

• Load Q&A JSON documents from the docs/ folder
• Embed and index them using Chroma and OpenAI
• Run a Streamlit app that answers questions using vector search and LLM fallback

---

1. Prepare JSON Q&A Files

Create a docs/ directory and add .json files structured like this:

[
  {
    "question": "What is AI?",
    "answer": "Artificial Intelligence is the simulation of human intelligence."
  },
  {
    "question": "What is Python?",
    "answer": "A programming language."
  }
]

2. Index the Documents

Run the document loader to embed and store the data:

poetry run python -m support_agent.scripts.docs_loader

This will:

• Read all .json files from docs/
• Create Q:/A: pairs
• Embed them using OpenAI
• Save them to a persistent Chroma vector store in vectorstore/

3. Run the Application

Launch the Streamlit interface with:

poetry run streamlit run src/support_agent/agent.py

Then open the app in your browser to begin asking questions.

4. Test Coverage

The application has ben tested with 100% code coverage

You can run all tests and view coverage:

poetry run pytest --cov=support_agent --cov-report=term-missing

Project Structure

support_agent/
├── agent.py                  # Main agent routing logic
├── fallback.py               # Fallback logic using OpenAI API
├── tools/
│   └── retriever.py          # Chroma-based vector search
├── scripts/
│   └── docs_loader.py        # JSON to vectorstore loader
├── assets/
tests/
├── test_agent.py
├── test_fallback.py
└── test_retriever.py
docs/                         # Your source Q&A JSON documents
vectorstore/                  # Persisted vector index

Environment Setup

Make sure to set your OpenAI key:

export OPENAI_API_KEY=your-key-here

Or add it to a .env file in the project root:

OPENAI_API_KEY=your-key-here

License

MIT