Data & AI Meetup #17

When your database stores data in columns and your client wants columns back, why is everything on the wire still rows and text? Javier will talk about how they designed a new wire protocol from scratch: binary, columnar, and bidirectional. He will share the protocol design decisions and their trade-offs. Great stuff! Speaker: Andrey Lukyanenko (Meta) Talk: AI-driven participation in Kaggle competitions Can you hand a Kaggle competition to AI agents? Mostly yes, but they still need direction. Andrey walks through real competitions where agents wrote code, ran long experiments, and achieved strong results, while also making false assumptions, chasing dead ends, and optimising the wrong signals. Andrey will share how multi-session memory, reusable skills, validation checks, and human judgement turn capable agents into a reliable research workflow. Speaker: Alexy Golev Talk: When the Bug Looks Like Success: Reliability Patterns for Multi-Agent AI Systems A distributed-systems lens on multi-agent AI/LLM agent architectures: the failure modes look new, but most are classic distributed-systems problems wearing an AI costume.Why an agent "succeeding" (producing plausible, confident output) VS an agent "working correctly" are not the same thing.Core failure patterns. Reliability patterns borrowed from established backend/distributed systems engineering and how they map onto multi-agent pipelines. Audience takeaway: a practitioner checklist of "boring" engineering practices that make agentic systems trustworthy, not just impressive in a demo. Please aim to arrive at 6pm. The talks will start around 6.30pm (give or take a few mins) and there will be pizza and refreshments as always on arrival :) Look forward to seeing you all soon!