
Stop Chatting with Your Docs — Architecting a Local-First AI Mind Coach in TypeScript with Gemma & Gemini Argon | Adrian Romanski @ DevFest Warsaw
About the event
Adrian Romanski will be speaking at DevFest Warsaw, one of the region’s premier community-driven tech conferences. DevFest brings together developers, engineers, and tech leaders to share insights on the latest advancements in software engineering, AI, and cloud technologies.
In his upcoming workshop, Adrian is challenging the current standard of AI integration. Instead of a passive tool that simply summarizes documents and stunts active learning, he will show attendees how to engineer a demanding, two-tier Socratic coach in TypeScript. By combining edge inference with zero-knowledge privacy patterns, his session explores how to use AI to stress-test your engineering knowledge securely.
Abstract
Most AI integrations for developer notes boil down to a simple "chat with your docs" wrapper. While convenient, this model creates a dangerous illusion of mastery: when AI serves ready-made summaries on demand, active thinking stops and deep learning fades away.
What if we inverted this architecture in TypeScript turning our private knowledge base into a demanding Socratic coach with the absolute privacy of an offline notebook?
In this approximately 45-minute workshop, we will dissect the architecture of a two-tier cognitive sparring partner that stress-tests your engineering knowledge without ever sending a single private file to the cloud:
Local Gemma (The Offline Sparring Partner): Running entirely on your machine against raw Markdown notes, it acts as a relentless coach on trains or flights driving Socratic dialogue, banning industry buzzwords, and forcing you to explain complex mechanisms in your own words using the Feynman technique.
Gemini Argon (The Zero-Knowledge Judge): Stepping in when connectivity returns, we leverage Argon within a strict LLM-as-a-Judge pattern via the @google/genai SDK. Using a blind claim-extraction protocol, Argon never sees your underlying notes, only the generated challenge and your human response ruthlessly exposing logical fallacies, shallow hand-waving, and gaps in your mental model before scheduling your next review.
Join this session to learn how to combine edge inference, zero-knowledge privacy patterns, and frontier reasoning with Gemini Argon to build tools that upgrade rather than replace your engineering brain.
Meet the speaker

Adrian Romanski
- Google Developer Expert
- Trainer

