RSQKit Chatbot

A Retrieval-Augmented Generation (RAG) chatbot connected to the RSQKit. Built with Streamlit for the web interface, ChromaDB for vector storage, Ollama for local model serving, and supports multiple AI providers via OpenAI-compatible APIs.

📌 Features

• Retrieval‑Augmented Generation (RAG) Chatbot
  Leverages document retrieval and generative AI to provide accurate, context‑aware answers.

• Multi‑Provider AI Support
  Compatible with Ollama and any OpenAI‑style API for both language models and embedding services. Set default models, and API credentials without modifying the code using the provider_config.yaml file.

• Hybrid Search Engine
  Integrates BM25 keyword search with vector‑based similarity to ensure precise and relevant results.

• Streamlit‑Powered Interface
  Interactive web app featuring chat history, source traceability, and real‑time sidebar controls.

• Modular Architecture
  Clean separation of ingestion, configuration, and application logic for easy extension and maintenance.

📦 Installation

Prerequisites

Before you begin, please make sure you’ve installed all of the following. Ollama is only needed if you plan to run models locally. If that’s the case and you haven’t installed it yet, head over to the official docs and follow the platform-specific instructions: https://ollama.com/

• Python 3.10 or higher

• Ollama
  After installation, confirm it's working by running:

    ollama --help

• API Key from an AI provider that supports the OpenAI API protocol (required for using remote models)

• Embedding Model for semantic search (e.g., bge-m3 with Ollama). Install it locally with:

  ollama pull bge-m3

• LLM for text generation (e.g., deepseek-r1:14b with Ollama). Install it locally with:

  ollama pull deepseek-r1:14b

• Reranker (optional) — improves ranking accuracy of search results.

---

Configuring AI Providers

To integrate an AI provider, update the provider_config.yaml file with the provider's details. Below is an example configuration format:

new_provider:
  name: "Example Provider"
  base_url: "https://api.exampleprovider.com/v1"
  base_url_vision: "https://api.exampleprovider.com/v1/chat/completions"
  rerank_url: "https://api.exampleprovider.com/v1/rerank"
  api_env_var: "API_KEY_EXAMPLE_PROVIDER"
  fall_back_provider: "ollama"
  supports_embedding: true
  supports_reranker: true
  models:
    default_embedding: "example-embedding-model"
    default_reranker: "example-reranker-model"
    default_llm: "example-llm-model"
    default_vision: "example-vision-model"

Steps

1. Clone the repository
   Open your terminal and run:

   git clone https://github.com/EVERSE-ResearchSoftware/RSQKit-chatbot.git
   cd RSQKit-chatbot

2. Set up a virtual environment

   Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies

   With the virtual environment activated, install the required packages:

   pip install -e .

4. ⚠️ Configure environment variables

   Create a .env file and populate it with your API keys and other required settings as referenced in provider_config.yaml.

🚀 Usage

1. Make these Bash Files Executable

chmod +x download_rsqkit_files.sh ingest_rsqkit_files.sh run_app.sh

2. Download and Ingest RSQKit Files

./download_rsqkit_files.sh  # This will take a moment. Skip if you already have the files in the directory `rsqkit_markdown`.
./ingest_rsqkit_files.sh    # This will also take a moment. Ensure you have an embedding model available.

This will download RSQKit files and ingest them into ChromaDB for retrieval.

3. Run the Chatbot App

./run_app.sh

Opens a Streamlit web interface at http://localhost:8501.

📁 Project Structure

rsqkit-chatbot/
├── app.py                              # Main Streamlit app
├── chroma_data_ingestor.py
├── core_utils                          # Containing utility functions (data processing, retrieval)
├── directories.yaml
├── download_rsqkit_files.sh
├── ingest_rsqkit_files.sh
├── llm_provider_tools.py
├── llms                                # Contains LLM chat functions for OpenAI protocol
├── pages                               # Contains the other pages of the app - Streamlit protocol
├── prompt_templates
├── provider_config.yaml                # ⚠️ Where you setup AI provider, API_KEY_VAR, LLM, embedding, etc
├── pyproject.toml
├── pytest.ini
├── README.md
├── rsqkit_scrap.py
├── rsqkit_markdown                     # Contains markdown input files (not tracked)
├── run_app.sh
├── settings.py
├── static
├── task_modules
├── templates
├── tests
├── ui
├── uv.lock
└── vision                              # Contains function for Vision Language Models

📝 Contributing

We welcome contributions of all kinds!

• Bug Reports: If you encounter an issue, please open a GitHub issue with a clear and detailed description.
• Feature Requests: Have an idea for an enhancement? Open a discussion or submit a pull request with your proposal.
• Code Contributions: Fork the repository, make your changes, and submit a pull request. Make sure to follow any existing coding conventions and include relevant documentation or tests where applicable.

📄 License

This project is licensed under the Apache-2.0 License.
See the LICENSE file for full details.

🙏 Acknowledgments

• This project is financially supported by EVERSE, IJCLAB, IN2P3, and CNRS
• We extend our gratitude to the teams at AI4EOSC, Albert API and RAGaRenn for generously providing their AI resources during the development of this application.
• The project draws inspiration from Retrieval-Augmented Generation (RAG) and hybrid search strategies in modern AI engineering.
• The application is built using tools like Streamlit, ChromaDB, and Ollama.