Collections

Best Chain of Thought Episodes on RAG & Retrieval

Getting the right context to the model: chunking, vectors, graphs, and what breaks at scale.

Retrieval is the least glamorous part of an AI system and the one that decides whether it works. The arc across these five is a field discovering that the first version (chunk the documents, embed them, search by similarity) was the easy 80%, and that the remaining 20% is data architecture, not model choice. Ordered newest first.

  1. 1 EP 55 52 min Transcript Hallucinations Are a Data Architecture Problem | Sudhir Hasbe, Neo4j Sudhir Hasbe, Neo4j Neo4j’s Sudhir Hasbe walks the standard pipeline end to end, chunk to embed to similarity search, then argues where it hits a ceiling and what a knowledge graph adds on top. The clearest explanation of the mechanism on the show.
  2. 2 EP 53 59 min Transcript Agent Memory: The Last Battleground in the AI Stack | Richmond Alake, Oracle Richmond Alake, Oracle Richmond Alake’s finding from building agent memory at Oracle: “vector search was not all you need… It was a very key component of what you needed, but it wasn’t all you need.” Retrieval strategy is what the memory layer rests on.
  3. 3 EP 49 54 min Transcript How Intercom Cut $250K/Month by Ditching GPT for Qwen Fergal Reid, Intercom Fergal Reid on running what he believes was one of the first production RAG deployments, launched on GPT-4 launch day, and what three years of iterating the chunking and context pipeline behind Fin actually taught them.
  4. 4 EP 43 53 min Transcript Beyond Transformers: How Liquid AI Is Rethinking LLM Architecture | Maxime Labonne Maxime Labonne, Liquid AI Maxime Labonne on the retrieval half nobody demos: the embedding models themselves, and why Liquid AI built a late-interaction ColBERT retriever rather than reaching for classic single-vector embeddings.
  5. 5 EP 8 35 min Transcript AI Infrastructure & the Evolution of RAG | Weaviate's Bob van Luijt Bob van Luijt, Weaviate Weaviate’s Bob van Luijt on RAG’s original one-directional shape of query, retrieve, generate, and the moment agents turned it into a feedback loop where the model decides what to fetch next.

More collections

All collections →

New here? Start with five → Browse by topic →