Basic Custom Copilot

Inspiration

The inspiration behind this project was to explore the possibilities of customizing and fine-tuning language models for specific tasks and domains, as demonstrated in the hackathon's guide.

What it does

This project involves creating a custom copilot that streamlines the dataset provided by the guide to create a context where customers can get the store-relevant data in human language.

How I built it

I followed the hackathon's guide to fine-tune gpt-4 on a custom dataset.

Challenges we ran into

I didn't use the bicep template to provision resources in Azure, and because of this I had difficulties in integrating the frontend with the backend. The frontend wasn't registering the backend API endpoint variable, so I had to do a workaround.

Accomplishments that I'm proud of

I'm proud of learning about how to utilize LLMs for specific needs and how to use Azure services.

What I learned

• vCore-based Azure Cosmos DB for MongoDB
• Container Apps
• Azure OpenAI API
• chatGPT models and their differences
• Embeddings, RAG and vector search
• Fine-tuning chatGPT for specific use-cases
• Deploying to Azure Container Apps and Azure Web Apps

What's next for Basic Custom Copilot

I want to build on this to make a copilot for the second phase.

Built With

• azure
• chatgpt
• cosmosdb
• openai-api
• python