nitya / contoso-web Goto Github PK

This is the Contoso Outdoors Company Website shown at Microsoft Ignite. It uses assets created by DALLE-3 and GPT-4. It is built with Next.js and Tailwind CSS.

First, run the development server:

npm run dev
# or
yarn dev
# or
pnpm dev
# or
bun dev

If you open http://localhost:3000 with your browser you should get the following result:

Click on the chat button indicated above to start a chat session. There are three different chat types you can try:

• Regular Chat using prompt flow (you can set that up using the Contoso Chat repo). This is the default chat type.
• Grounded chat using the "Add Your Data" feature inside of the Azure AI Studio playground (if you follow the instructions in the Contoso Chat repo, you will have this set up already). This can be enabled by adding ?type=grounded to the URL. For example, http://localhost:3000/?type=grounded
• Visual Chat - this is the same UI that was shown at the Microsoft Ignite conference. This can use either video capture or an image upload. This can be enabled by adding ?type=visual to the URL. For example, http://localhost:3000/?type=visual. If you want to use video capture, you can use ?type=video instead. For example, http://localhost:3000/?type=video. (Instructions for how to set this prompt flow up are forthcoming as soon as GPT-4 Turbo with Vision is released).

In order to use the Contoso Outdoors website, you will need to set up the following endpoints in a .env file in the root of the project:

CONTOSO_SEARCH_ENDPOINT=<YOUR_SEARCH_ENDPOINT>
CONTOSO_SEARCH_KEY=<YOUR_SEARCH_KEY>
CONTOSO_AISERVICES_ENDPOINT=<YOUR_AI_SERVICES_ENDPOINT>
CONTOSO_AISERVICES_KEY=<YOUR_AI_SERVICES_KEY>
PROMPTFLOW_ENDPOINT=<YOUR_PROMPTFLOW_ENDPOINT>
PROMPTFLOW_KEY=<YOUR_PROMPTFLOW_KEY>
VISUAL_ENDPOINT=<YOUR_PROMPTFLOW_VISUAL_ENDPOINT>
VISUAL_KEY=<YOUR_PROMPTFLOW_VISUAL_KEY>

If you follow the Contoso Chat repo, you will have all of these endpoints set up already. If you want to use the Visual Chat feature, you will need to wait until GPT-4 Turbo with Vision is released.

As part of this site we have added debug statements to the console to see the responses. For the grounded chat, you will see the following:

For the standard prompt flow chat you will see the following:

This is useful for debugging purposes and shows you what is being sent to the individual endpoints.

Couple of things I would like to do with this repo:

1. Change to streaming chat output instead of waiting for the entire response to come back.
2. Any other ideas? Let me know!