Home
Igor Pavlov portrait

AI Engineer · LLM Systems · Backend

Igor Pavlov

AI product and LLM systems engineer

[email protected] t.me/beastia linkedin.com/in/gophan igor-pavlov.ru

I build AI products: from architecture to interface, from concept to live client data. 13 years of backend practice means I think about cost, reliability, and scale — not just model output quality. I've been CTO at a startup and am choosing an engineering role deliberately: I want to stay in the code and build systems, not manage communications.

AI agentUsers ask questions to their own data and get answers — shipped to production
Model selection by dataCompared 5 LLM providers, measured +18.8% improvement in composite quality score
AI Knowledge BaseStatic docs → live AI support tool actively used by the team (Uzum, 2022–2025)
13+ yearsRuby, Elixir, Python — fintech, distributed systems, production scale

AI experience

Deeplens

CTO / Lead AI Engineer · 2025–2026

AI startup: SaaS platform for product teams — turns customer feedback and reviews into structured insights with recommendations.

  • Built the platform from zero to live client data — architecture, LLM pipeline, and interface owned end-to-end alone.
  • Built a strategy across five LLM providers (OpenAI, Yandex GPT, Qwen3, DeepSeek, GigaChat) with automatic fallback; primary model chosen by benchmark: +18.8% quality score improvement.
  • Designed the feedback analysis pipeline: vector embeddings + pgvector, RAG context, two-phase generation with hallucination check and per-insight quality scoring.
  • Shipped a conversational AI agent on MCP protocol: users ask questions to their own data in dialogue and get instant answers.

Uzum · Tashkent

Unit Tech Lead · 2022–2025
  • Transformed static documentation into an AI-assisted knowledge base—a live tool actively used by customer support and engineering teams.
  • Introduced AI-assisted tooling into team workflows to automate repetitive processes and improve delivery speed.
  • Backend architecture for fintech systems with zero failure tolerance: Elixir, PostgreSQL, Redis, ClickHouse—40,000+ cards issued through services I designed.

Backend foundation

2011–2022

Ruby, Rails, Elixir, Python, distributed systems, NATS, ClickHouse, infrastructure. Open source: NATS clients for Ruby and Elixir, a ClickHouse package, prerender.io fork.