Jettro Coenradie

Software architect, search relevance engineer and Gen AI practitioner — by his own account he started as a software developer before moving into search, Gen AI and LLMs. Writes hands-on build walkthroughs. Authored the Embabel publication’s account of putting DICE to work in a chatbot — extracting user preferences from conversation as scored propositions and injecting them into later turns alongside RAG.

His framing of the problem is the grounding gap: an LLM has no common ground with the user’s question, and a system that learns about the user from observation can manufacture that common ground rather than asking for it.

Articles in this vault

• Agents That Extract and Use Preferences from Conversations — building on Embabel DICE; the GUM lineage, the proposition pipeline, and extraction moved off the response path via an event listener

Topics

• DICE · Agentic Memory · Context Engineering · Personalization

Related

• Embabel · Embabel DICE — the framework and module he builds on
• Andrew Kornilov — independently implemented the same design as django-dice