This example demonstrates how to create a ChatGPT-style application using the Retrieval Augmented Generation (RAG). It leverages Azure's OpenAI Service to access the ChatGPT model (gpt-3.5-turbo) and employs Azure Cognitive Search Vector Search for data indexing and retrieval.
The chatbot can analyze and comprehend the details of your uploaded images. This enables you to interact with the chatbot as if you were conversing with someone who has viewed the same photos.
To deploy the application, please ensure that you have the following dependencies installed on your machine.
- This repository is still in early development.
- Your photos are currently stored in a public Azure Blob Storage container. Please do not upload any sensitive data.
- Clone the repository and navigate to the project root
- Run
azd auth login - Put the photos you want to index in the
./datafolder - Run
azd up - Make sure to select a region with quota available for both GPT-4-32k & GPT-4-Vision. (e.g. 'switzerlandnorth') as the region.
The current indexing process is suboptimal. During my evaluation, the system could only handle an approximate rate of 300 images hourly.
- Improve indexing performance
- Ability to query based on image metadata (e.g. location)
- Bot Framework Support (In progress)
- Reduce Hallucinations / Prompt tweaks