A detailed walkthrough of an initial approach to semantic chunking for Retrieval Augmented Generation over video timestamps.

Retrieval Augmented Generation (RAG) systems are powerful, but their performance heavily relies on the quality of context provided to the Large Language Model (LLM). When dealing with extensive content like video tutorial transcripts, naive chunking can lead to fragmented, irrelevant, or incomplete information, ultimately degrading the user's experience. This article presents the first-iteration of a practical chunking strategy implemented in PsTuts RAG project as a part of the learning path toward LLM engineering (s/o AI Makerspace ). I'll detail how we combine semantic chunking with timestamp alignment to tackle these challenges, offering a method to create contextually rich and accurately timed chunks fro