About London Reactor Meetup
Microsoft Reactor London Meetups are a community of like-minded engineers, developers, professionals and students, allowing you to connect with people, skills, and technology to enhance your career or personal learning. Our regular meetups are an opportunity to hear from leaders in the industry about challenges, tooling and best practices. If you are interested in speaking or want to suggest a talk you’ve heard before, please do let us know (no vendor / recruitment / sales pitches please).

Agenda

6:00 PM - Arrivals: Registration and welcoming of guests
6:30 PM - 6:40 PM - Welcome Address
6:40 PM - 7:00 PM - Use an LLM on the tube with offline AI by Jim Bennett
7:10 PM - 7:40 PM - Lessons from setting up RAG systems in production to search over internal data and automate customer support by Victor Naroditskiy
7:40PM - Closing, snacks & networking

Session Descriptions

Use an LLM on the tube with offline

We've all got used to using LLMs in our developer workflow - from asking ChatGPT what tools and libraries to use, to getting GitHub copilot to generate code for us. Great when you are online, but not so useful when you are offline, like on the tube or in a plane with no WiFi. But what if there was another way?

In this session, Jim will introduce offline LLMs. We'll look at how you can run LLMs locally, such as Phi-2 from Microsoft, and add these to your developer workflow. We'll compare the performance of offline vs online, both the speed and quality, but also touch on privacy and other considerations. We'll also look at hardware requirements as we don't all have the latest GPUs to hand.

Lessons from setting up RAG systems in production to search over internal data and automate customer support

Everyone has been crazy about Gen AI lately and yet few companies can say they derive measurable business value from Gen AI based solutions. I will go over two use cases and the journey to get them from an out-of-the-box RAG pipeline that kinda works in a demo to production quality systems that create value. The first use case is search over technical content such as documentation and developer forums. The second use case is automating ticket-based customer support.