Ayaal

Santaev

I'm an AI/LLM engineer who needs to understand things all the way down - I don't like shipping a mechanism until I know why it works. That pull toward hard, unfamiliar problems has taken me from control systems for nuclear power plants at Rosatom, through Solana and Cosmos infrastructure, to retrieval systems, agents, and MCP tooling today. The clearest example of how I work is my retrieval research (github.com/Galiusbro/galtonian-context-retrieval): I designed a new method, benchmarked it properly - multi-seed runs, bootstrap confidence intervals - and published the results even though the baseline won on the main metric. A result you can't trust is worse than no result. The same habit runs through my production code: strict typing, CI, and tests on the paths that can hurt. Currently building an AI research platform at NUAH/Universa. Python, TypeScript, FastAPI, PostgreSQL / Neo4j / Qdrant / Redis - and Rust or Go when the problem asks for it