: The debut of DolphinGemma
Post Content
We are thrilled to introduce the first episode of new Microsoft Engineering Video Series: AI in Action!
In this series, we will be deep diving with Microsoft engineers on how they are building AI applications here at Microsoft. We’ll cover many different topics in this series, starting with one of the most common updates that developers are making with AI: implementing retrieval-augmented generation (RAG) into applications. In this video we will learn directly from the Microsoft Engineering team on how they used RAG to develop the Ask Learn function on Copilot for Azure at scale! From discussing chunking strategies to how to avoid hallucination, deep dive into how Microsoft builds their own AI.!
Get Started: AI for Developer Hub: https://aka.ms/AIforDeveloper
Let’s connect:
Brian Steggeman | LinkedIn
Eric Imasogie | LinkedIn
Tianqi Zhang | LinkedIn
Frank Boucher | LinkedIn
00:10 – Introduction
00:49 – Demo
03:25 – What is RAG?
07:30 – What part of this product is being used from Microsoft?
09:50 – Why do you think it was necessary to create this plugin for Ask Learn?
11:35 – What was so special about this project?
13:01 – What challenges did you face building this tool?
17:10 – If there is a bad response, how do you improve it for next time?
18:24 – Were there any surprises you faced and how did you deal with them?
21:25 – What steps do you take to clean and preprocess data before it’s ingested into the system?
24:08 – There is a lot of information on Microsoft Learn, how did you break it down, what was your strategy?
27:25 – How did you manage data changes consistently?
29:49 – How do you keep a balance between a fast and good answer?
32:33 – How do you know that Ask Learn is doing well and how do you monitor it?
34:34 – How do you avoid hallucination and ensure safety?
36:22 – Security Concerns
37:28 – What advice would you give to developers who are new to building RAG-based applications?
39:33 – What are some mistakes/traps to avoid?
43:20 – Conclusion Read More Microsoft Developer
