Part 5/11:
Types of data: Raw logs, metadata, and user preferences.
Advanced search techniques: Use of Azure Search, full-text indexing, and vector-based embeddings to perform semantic searches.
He proposes organizing conversation data into knowledge graphs—visual representations linking entities and concepts—enabling complex summarization, retrieval of relevant past conversations, and continuous model training.
However, Shapiro candidly notes that much of the AI’s output seems superficial because it doesn’t necessarily require deep long-term memory—it's more about how the data is leveraged, not just stored.
Pitfalls and Challenges
A significant part of his discussion revolves around potential pitfalls: