AI engineer in Kazakhstan (UTC+5). I build LLM pipelines with automated quality checks and run them in production: generative video, multimodal LLM-as-judge, RAG over code. I also build playable ads in TypeScript. MSc from MIPT, PhD candidate in computer vision.
- playable-ads: two playable ads, a squad runner and a merge tower defense, on Three.js and PixiJS, with every mesh, texture and sound generated in code. One code base builds 3 A/B variants for 8 ad networks, each tested against that network's SDK stub. Play them in the browser
- code-rag-mcp: local hybrid code search for coding agents, measured on 70 questions with held-out sets
- narration-pipeline: writes narration for explainer videos with fact statuses and about 25 deterministic checks, then cuts it into frames with an exact dynamic program
- procedural-films: short films where every frame and sound is computed in plain JavaScript and Canvas
Open to remote roles. laido59@gmail.com · t.me/deselection

