Event Timeslots (1)
Track 3 – 2024
-
Presenter: Trevor Bennett
Abstract: "Have you ever interacted with a large language model, only to realize that its answers are generic, and possibly even made up? This is an evergreen issue in cases where answers to the questions exist, but the model lacks the ability to find them, because it was never trained on the data in question in the first place. We have the data (somewhere), and we have the model, how do we make them talk?
Retrieval augmented generation is a technique wherein you take a user's question, enhance it for the purposes of a document query, and then use those same documents to contextualize the user's question to the model.
We're going to build a workflow that does exactly that. In this talk we will walk through the architectural decisions and design constraints that lead to us using LangChain, Amazon Bedrock, and Open Search to build out a retrieval augmented GenAI pipeline. We will conclude with a comparative demonstration of results and a discussion of lessons learned."
AWS Services: bedrock, bedrock agents, lambda, s3, dynamodb, open search
Audience: Beginner
angelo