Solution Architecture of KymChat Conversational AI Assistant

If you are curious what it takes to deploy an AI solution

As of 2024, you probably have already used a lot of BBC by this time. By BBC, I meant Bard, Bing, and ChatGPT. I’m sure you have come across or used all 3 main use cases of AI:

• Generate e.g. copywriting, brainstorming

• Reduce e.g. summarisation

• Transform e.g. translation, refactoring, data transformation etc.

Look at these two cuties exchanging emails

Recently, a colleague shared on their custom GPT called KymChat, built using Azure OpenAI service, with the main advantage over ChatGPT as it’s highly secure and private, with all data remaining under the entity's control. It got an endorsement on Microsoft’s website, and I’m curious what it is all about. Let’s dig in.

Technology Stack:

• Generative AI Models: GPT-3.5, GPT-4 (through Azure OpenAI Service)

• Database: Azure Cosmos DB for MongoDB vCore (with vector search capabilities)

• Backend:

  • Code stored in Azure Container Registry Docker image

  • Azure App Service to deploy containerized web app based on the Docker image

• Other Services:

  • Azure App Service (for automatic scaling)

  • Azure Container Registry

  • Microsoft 365 productivity files

  • KPMG Prompt Confidence Index (internal tool)

If you are experienced in the AI/engineering field, the above is enough, however, if you are relatively new, here are a few more details on the components:

• Azure Cosmos DB for MongoDB vCore: Stores and retrieves data, enables fast and accurate search using vector search.

• Azure OpenAI Service: Provides access to GPT-3.5 and GPT-4 models.

• Azure App Service: Hosts the KymChat web app and allows for automatic scaling based on demand.

• Azure Container Registry: Stores the Docker image used to deploy the KymChat web app.

• Microsoft 365 productivity files: Source of data for KymChat

Happy chatting!