Blackboard Learn RAG Chatbot Proxy (Ollama)

This project provides a Node.js proxy service that enables you to run a chatbot using open-source LLMs (via Ollama), but restricts the chatbot's knowledge to only the content found on the Blackboard Learn Help site. It implements a simple retrieval-augmented generation (RAG) pipeline: user questions are matched to relevant website content, and only that content is provided as context to the LLM for answering.

Features

• Retrieval-Augmented Generation (RAG): Answers are based only on content from Blackboard Learn Help, not the model's general knowledge.
• Local LLMs: Uses Ollama to run models like Llama 2, Mistral, or Phi-3 on your own machine.
• Simple REST API: POST /chatbot with a question, get a focused answer.
• Easy Content Updating: Re-crawl the site anytime to refresh the knowledge base.

Installation & Setup

1. Clone the Repository

git clone https://github.com/ShannonH/chatbot-middleware
cd chatbot-middleware

2. Install Dependencies

npm install

3. Install and Set Up Ollama

• Download and install Ollama from https://ollama.com/download
• Pull a model (e.g., Llama 2):

  ollama pull llama2

• Make sure Ollama is running (it usually starts automatically):

  ollama serve

4. Crawl Blackboard Learn Help Content

This will download and extract content from the help site for use by the chatbot.

npm install axios cheerio
node crawl_blackboard_learn.mjs

• This creates blackboard_learn_content.json in your project directory.
• You can adjust the crawl limit in the script if you want more or fewer pages.

5. Start the Proxy Service

node ollama-proxy.mjs

• The service will listen on port 4000 by default.

Usage

Send a POST request to /chatbot with a JSON body:

{
  "message": "How do I create an assessment with different question types in Blackboard?"
}

The response will be:

{
  "reply": "...answer based only on Blackboard Learn Help..."
}

How It Works

1. The proxy loads the crawled content from blackboard_learn_content.json.
2. When a question is received, it searches for the most relevant content chunks.
3. It builds a prompt for the LLM that includes only those chunks and instructs the model to answer using only that information.
4. The LLM's response is returned to the user.

Customization

• Change the Model: Edit ollama-proxy.mjs to use a different model (e.g., mistral, phi3).
• Improve Retrieval: For better search, consider using embeddings and semantic search libraries.
• Expand/Refresh Content: Re-run the crawler script to update the knowledge base.

Troubleshooting

• If you get No relevant Blackboard Learn content found for your question., try re-crawling or adjusting the search logic.
• If you see errors about missing modules, ensure all dependencies are installed.
• Ollama must be running and the model must be pulled before starting the proxy.

License

MIT

---

Questions or suggestions? Open an issue or PR!