Dylan Couzon leads Developer Relations at Qdrant, the open-source vector search engine, where he helps developers build retrieval and memory into their AI applications. He teaches the DeepLearning.AI short course Building AI Assistants with On-Device Memory, made in partnership with Qdrant, and spoke on AI memory at AI Engineer World’s Fair 2026 in San Francisco.

Before Qdrant, he worked in developer relations at Arize AI, where evaluating LLM applications was his core focus. He still builds that way: he ships retrieval systems, measures whether they help, and has demoed robot memory running on Qdrant Edge on hardware as small as a Raspberry Pi.

He learned to code 18 years ago to hack on video game emulators and has been taking side projects too far ever since. He’s based in New York.

// 02 / The session

Simulated fly brain, persistent trauma

Abstract

The internet keeps tormenting simulated fly brains. Dylan is giving one a memory so it can remember the bad times, then testing whether past experience helps the same brain avoid unfamiliar hazards and survive longer.

Along the way you’ll learn how memory helps agents keep useful experience across sessions. You’ll use search and retrieval to build a memory system that lets an agent recall what happened before and apply it to new challenges. You leave with a working system that learns from experience, plus reusable code and tests for your own applications.

What you’ll learn

Use retrieval to give agents memory across sessions

Apply past experience to unfamiliar situations

Test whether memory improves an agent’s outcomes