Jeng Y.

Hi folks, Just wrote a piece about contextual optimization. As LLMs turn data processing into semantic processing, we believe ETL processes will turn into “ECL” (“Extract-Contextualize-Load”) processes. We explore what “ECL” could mean in the context of LLM information retrieval / RAG processes look like, and what the history of ETL can teach us about the development of the ECL industry, and the supporting tooling that would likely emerge. https://medium.com/enterprise-rag/improving-llm-information-retrieval-etl-to-ecl-extract-contextualize-load-12a4ac259faa

There’s been a bunch of questions thrown up about Gemini 1.5’s implications for RAG. I’m pretty bullish that Gemini 1.5 will accelerate RAG systems to production. Gemini 1.5’s 40% failure rate in multiple-needle recall, as well as lack of meaningful ways to debug the black-box of an LLM will mean more auditable systems like RAG will give back control to developers. Further, at end-state, being able to optimize for Accuracy, Costs and Latency are also issues that mean that the last-mile of optimization will always be RAG-based. Curious to learn more about how others think! https://medium.com/enterprise-rag/why-gemini-1-5-and-other-large-context-models-are-bullish-for-rag-ce3218930bb4

Recursive or iterative retrieval is one of the more exciting approaches to being able to create self-learning RAG systems that are able to learn over time, and to deeply explore unstructured data. This LLM & Knowledge Graph exploration can be done either autonomously, or in directions that you point the knowledge graph at. We categorize and go into the 3 types of recursive retrieval techniques we’ve identified so far - Page-Based, Information-Centric, and Concept-Centric iterative retrieval in advanced RAG. Check it out here: https://medium.com/enterprise-rag/advanced-rag-and-the-3-types-of-recursive-retrieval-cdd0fa52e1ba WhyHow.AI (RAG Infra) provides context optimization tooling for Vertical AI devs, using the power of knowledge graphs. If anyone wants to chat about RAG and/or knowledge graphs, hit me up!