Two decades building backend, platform, AI, and product systems. More than a decade leading engineering teams through ambiguity, scale, delivery pressure, and technical change.
My work with AI and machine learning goes back to around 2010, before the current GenAI wave. Over the years, that has included a TensorFlow + IoT irrigation system for high-end farming in 2019, and an LLM-based engine in 2024 that became the initial core of a major product initiative.
Today I lead Search Backend at Kiwi.com, one of travel's highest-volume and highest-stakes domains. Alongside that, I build AI-native systems for knowledge, coordination, engineering workflows, and writing.
I care about the same things across all of this work: strong fundamentals, honest tradeoffs, pragmatic architecture, and systems that remain understandable as they scale.
- Search backend at scale. Reliable, high-volume services where latency and correctness both matter.
- AI-native engineering. RAG, agents, MCP, and LangGraph applied to real workflows rather than demos.
- Knowledge systems. Connecting documents, code, and memory into something a team can actually use.
- Engineering leadership. Growing engineers, setting technical direction, and reducing friction in how teams ship.
- Compendium: a local-first knowledge synthesis system that turns books, papers, articles, and notes into a canonical Markdown wiki, then retrieves and composes answers from stable, citable pages rather than raw chunks.
- Ubongo: a personal, local AI mind with memory, mood-aware interaction, governed tool use, self-improving prompts, self-authored skills, and MCP connectivity.
- dandori: a post-Scrum coordination framework for AI-augmented engineering teams, licensed under CC BY-SA 4.0.
- Agentic-Systems-Patterns: reusable patterns for designing and building agentic systems.
- AI-parrot: a LangGraph-based agent that ranks news and turns it into a TTS podcast.
Most of my current work sits around local-first AI systems, knowledge retrieval, autonomous writing, engineering copilots, and multi-agent workflows. Some of it is public, some of it is still private while I shape the architecture, harden the workflows, and decide what is useful enough to share.
Python · TypeScript · GoLang · C# · PostgreSQL · OpenSearch · Qdrant · LangGraph · MCP · RAG