Build RAG Apps in YAML - Intel AI DevSummit, February 26, 2025

Описание: This session addresses accuracy limitations of traditional vector-indexing RAG for large, dynamic private data. It will explore how to build quality RAG applications for end users with easy-to-use RAG templates that can be deployed easily via docker, with the index of your choice. You’ll learn to optimize RAG for the best balance of cost, accuracy, and latency. We’ll showcase a quick, secure RAG service setup (CPU+GPU or CPU) with pre-built YAML templates and automatic data updates, improving document understanding and performance. You will see how it works in practice, with real-world querying of sales and marketing documents and slide decks.

Speaker: Szymon Dudycz, AI Developer
Speaker Bio: From building the core of Pathway's AI engine to crafting high-performance Monte Carlo simulations at Credit Suisse, Szymon knows how to make complex systems work. He's the wizard behind Pathway's YAML templates. He holds a PhD on approximation algorithms.

Timestamps:

00:00 Introduction and Speaker Introduction
00:59 Overview of Pathway and AI Pipelines
03:18 Pathway's Slide Search Service
07:29 Architecture and Technical Details
12:41 Live Demo: Slide Search in Action
25:06 Customizing Pipelines with YAML
33:57 Q&A Session
40:48 Closing Remarks and Next Session Preview