Process Briefings

Description

This project demonstrates an approach to prepare public data for Retrieval-Augmented Generation (RAG) using advanced AI models. The resulting output is designed to work with a local Large Language Model (LLM) setup on modest hardware. For context, check out Blog.md, which is the relevant part from a blog post on https://meanmedianmoose.ca.

The configuration utilizes:

• Ollama for model management.
• OpenWebUI as the front-end interface.
• nginx for secure communication.
• Docker for containerizing the application and nginx.

Data Source

The demo data is based on a slightly cleaned version of the Health Canada Deputy Minister Briefing Materials published by the Government of Canada.

Available Output

The goal is a tool that supports natural-language queries about the structure, operations, strategic goals, and challenges of Canada’s health ministry.

Three processed outputs are included in this repository:

• dmbriefing_qwen2.5-14b.json
• dmbriefing_qwen2.5-72b.json
• dmbriefing_llama3.3.json

Use dmbriefing_qwen2.5-72b.json for the best results. This file can be loaded into an LLM's knowledge base for optimal query responses.

---

Getting Started

1. Clone the Repository

git clone https://github.com/yourusername/process-briefings.git
cd process-briefings

2. Set Up Docker

Ensure Docker is installed and running on your system. Use the provided Dockerfile to build and run the necessary containers:

docker build -t process-briefings .
docker run -d -v $(pwd):/app process-briefings

3. Configure nginx

Set up nginx as an SSL proxy for secure communication. Ensure you have the necessary SSL certificates and modify nginx.conf as needed. You can refer to this sample configuration.

4. Start Ollama

Run the Ollama service to manage the AI models:

ollama start

5. Launch OpenWebUI

This command is pulled from the OpenWebUI Git repository: https://github.com/open-webui/open-webui. Check that URL for updates or the right docker command for your use case.

docker run -d -p 3000:8080 --gpus all --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:cuda

6. Run the Script

In the process-briefings Docker container shell:

python process_briefings.py dmbriefing.html dmbriefing_new.json

The first argument to process_briefings.py is the document to process, and the second argument is the output document.

7. Upload Documents

• Open your web browser and navigate to:

https://<your_device_ip_or_name>

• Navigate to the Knowledge section in OpenWebUI.
• Select + to create a new Knowledge base for your data.
• Upload one of the provided JSON files (e.g., dmbriefing_qwen2.5-72b.json) to populate the model's knowledge base.

8. Create a Custom Model

• Go to the Models section in OpenWebUI.
• Select + to create a custom model.
• Follow the prompts to configure and deploy your model. (For example, "Qwen2.5:14b" or any high-quality midsized LLM.)
• Add a tailored system prompt to guide the AI’s responses. You can find a sample prompt here

---

License

This project is yours to use under the MIT License. See LICENSE for more details.