About / Proof
I’m a hands-on AI engineer focused on production LLM systems: agent workflows, retrieval, evaluation, observability, and the control layer needed to make AI usable inside real products.
Over 10+ years, I’ve worked across fintech, ads, e-commerce, media, and web3 — turning ambiguous product problems into explicit AI systems with state, routing, evaluation, and operational boundaries.
Selected Highlights
Production AI systems
Owned end-to-end architecture and delivery across agents, retrieval systems, and evaluation layers.
Scale and outcomes
Shipped ML across 50+ products, supported 150M+ user-scale systems, and contributed to ~$20M revenue impact.
Current proof areas
Agent workflows, trust-critical retrieval, observability, and bounded automation under real product constraints.
Proof Highlights
- AI systems shipped across 50+ products
- Personalization systems at 150M+ user scale
- ~$20M revenue impact from ML systems
- Production LLM agents with <2s interaction latency
- Designing decision-driven AI systems with explicit state and control
Selected Experience
GriffinAI
Owned production LLM agent architecture and delivery, including TEA, Proposal Examiner, and supporting multi-agent / ops reporting systems.
Independent consulting
Built applied LLM systems for finance and e-commerce, including trust-critical knowledge retrieval and decision-support workflows.
Laboratory for Neuroscience and Human Behaviour at Sber
Led AI initiatives, shipped AI across 50+ products, and contributed to large-scale personalization and client-profiling systems.
Publications & Talks
Selected public work that shows longer-running technical depth and communication ability beyond current product execution.
Cognitive Architecture for Decision-Making Based on Brain Principles Programming
Procedia Computer Science · 2022
Background depth in cognitive architectures and structured decision-making.
Selected Public Writing
Who Are All These AI Agents? Part 1: Exploring the Basics with Examples
June 20, 2024
Early public writing on agent architecture, memory, planning, and tools.
Proposal Examiner launch post
December 2025
Strong public post tied directly to a shipped AI product.
Key External Links
- GitHub
- TEA (signup required)
- Proposal Examiner
- SQL content-ingestion walkthrough repository
How I Tend to Work
I usually start from user needs and product constraints, then make the workflow explicit: state, routing, tool boundaries, evaluation, observability, and the places where humans need to stay in control. I care less about surface novelty than about whether the system can be trusted, operated, and improved over time.