[Proposal]: Build Your Own Copilot Hackathon (Build Multi-Agent RAG Pattern apps using Azure OpenAI + AKS/ACA + Cosmos DB + Semantic Kernel)

Topic

We have developed a hackathon which is built around the Build & Modernize AI Applications reference solution to demonstrate how to build RAG Pattern applications and services using Azure OpenAI, Azure Cosmos DB, Azure AI Search. and Azure Kubernetes Service. The solution is built using the Semantic Kernel SDK as the AI orchestration layer on .NET 7.

The hackathon contents are fully developed and had multiple customer deliveries. The hackathon consists of 7 challenges and can be delivered in a one day format of 5 challenges. Or a 1.5 - 2 day format with all 7 challenges. The original source for this hackathon is here: Build & Modernize AI Apps Hackathon

We also have a comprehensive Coaches Guide with materials for coaches to prepare for and deliver the hackathon. This includes:

PowerPoint presentations for coaches to present before each challenge.
Written guidance for coaches for each challenge.
Short videos to walk through each challenge to help coaches prepare.
Final solutions for coaches to use to help guide attendees when needed.
a 90-minute video that goes deep into the underlying reference application this hackathon is built on to help them train themselves on this solution.

Abstract/Learning Objectives

Coaches and attendees for this hackathon will learn how to design, build and deploy RAG Pattern applications and services using Azure OpenAI, Azure Cosmos DB, Azure AI Search. and Azure Kubernetes Service. They will learn how to implement these solutions using Semantic Kernel as the orchestration layer.

Attendees will gain knowledge on the following services and concepts:

Services:

Concepts:

Below are the learning objectives for each challenges:

Challenge 1: Deploying the service in Azure for this solution and configure the application to debug locally.
Challenge 2: Configuring vector indexes. Data loading and vectorizing. Automating the process to make repeatable.
Challenge 3: Vectorizing user prompts, building system prompts, implementing conversational chat history.
Challenge 4: End to end testing for data loading and vectorizing. Prompt engineering and testing.
Challenge 5: Customizing with new data, entity management, Semantic Kernel memory types.
Challenge 6: Creating new Semantic Kernel Plug-ins and Functions.
Challenge 7: Integrating with Prompt Flow.

Delivery Date

Now

Authors

https://github.com/markjbrown
https://github.com/sandeepsnairms
https://github.com/ciprianjichici
https://github.com/hatboyzero
https://github.com/joelhulen

Other

These contents are fully developed and ready to go. We could use some assistance in getting this content into the WTH format.

Once here our plan is to keep this as the central place to publish this content. The underlying reference app and hackathon will continue be receiving regular updates to reflect changes in the capabilities for the Azure services used in this solution as well as Semantic Kernel SDK.

Code of Conduct

[X] I agree to follow this project's Code of Conduct

microsoft / WhatTheHack