RAG-chat-with-docs-chromadb

A Python-based Retrieval-Augmented Generation (RAG) project that enables question answering and summarization over user-provided documents using ChromaDB and Hugging Face embeddings, In simple term its a RAG based chat bot.

---

Overview

This project loads text documents, converts them into vector embeddings, stores them in ChromaDB, and uses a Large Language Model (LLM) to answer user queries or generate summaries based on the retrieved contextual data.

The document loading behavior is configurable through a YAML-based configuration system.

---

Project Structure

RAG-chat-with-docs-chromadb/
│
├── main.py
│   # Entry point of the application.
│   # Loads documents into ChromaDB (based on config)
│   # and invokes the LLM for summarization or Q&A.
│
├── src/
│   └── vector_db.py
│       # Handles vector database creation, embedding,
│       # and interaction with ChromaDB.
│
├── raw_data/
│   # Contains sample input documents for embedding.
│   # Currently supports .txt files only.
│
├── properties/
│   ├── vector-config.yaml
│   # Configuration file for vector database behavior.
│   #
│   # Example:
│   # load_into_vector: true
│
│   └── vector_config_loader.py
│       # Utility to load and parse vector-config.yaml.
│
└── README.md

---

Configuration

All vector database behavior is controlled via: properties/vector-config.yaml

Key Property

• load_into_vector
  • true → Loads documents from raw_data/ into ChromaDB
  • false → Skips document loading and uses existing vectors

This allows flexible control over when embeddings are regenerated.

---

Embedding Model

The project uses the following Hugging Face embedding model: sentence-transformers/all-MiniLM-L6-v2

This model provides fast and efficient sentence-level embeddings suitable for semantic search and RAG pipelines.

---

Supported Data Formats

• ✅ .txt files
• ❌ Other formats (PDF, DOCX, etc.) are not supported in the current implementation

---

How It Works

1. Configuration is loaded from vector-config.yaml
2. If load_into_vector is enabled:
   • Text files from raw_data/ are embedded
   • Embeddings are stored in ChromaDB
3. User queries are matched against stored vectors
4. Retrieved context is passed to the LLM
5. The LLM generates answers or summaries based on the context

---

Environment Configuration (.env)

The project supports multiple LLM providers. At least one API configuration must be provided in the .env file.
Based on the available configuration, the application will automatically initialize the corresponding LLM client.

An example configuration is provided in .env-example.

# OpenAI API Configuration

OPENAI_API_KEY=your_openai_api_key_here
OPENAI_MODEL=gpt-4o-mini

# Groq API Configuration (alternative to OpenAI)

GROQ_API_KEY=your_groq_api_key_here
GROQ_MODEL=llama-3.1-8b-instant

# Google Gemini API Configuration (alternative to OpenAI/Groq)

GOOGLE_API_KEY=your_google_api_key_here
GOOGLE_MODEL=gemini-pro

If multiple providers are configured, the application will select the appropriate LLM based on the internal configuration logic.

---

Requirements

• Python 3.8+
• LLM Provider API Key

---

Git url

https://github.com/Raghul-S-Coder/RAG-chat-with-docs-chromadb.git