Hey, I'm Yaroslav
Senior Full-Stack AI EngineerI build TypeScript SDKs, APIs, and product workflows for developer tools. Currently at Comet ML.
Let's build somethingI take messy expert workflows and turn them into products, typed SDKs, and APIs that people can rely on.
How I build
What I fall back on when the product is vague and the engineering is not.
Products, not tickets
I own problems end to end, from the database schema to the deploy. Software gets better when the engineer building it has felt the user's pain.
Speed is a feature
A narrow product slice shipped this week teaches more than a perfect spec next quarter. I write, build, and debug with AI assistants, and I still hold the work to explicit tradeoffs and software that actually runs.
Make the team faster
A senior engineer should make the people around them faster and better. I build tools and ship examples so the good decisions are easy to repeat.
Diagnose before you fix
Find the root cause and understand the system before you change anything. I do this for architecture decisions and for production bugs at 2 AM.
Stay uncomfortable
I've gone through medicine, four countries, three tech stacks, and open source in a protocol that barely existed. I do my best work at the edge of what I know.
What I'm building
I build product workflows and developer infrastructure for OPIK, an open-source LLM evaluation and observability platform at Comet ML with 22K+ stars. That covers the Ollie assistant workflows, the SDKs, and the debugging loop.
OPIK Platform
I own the OPIK frontend, built the TypeScript SDK from scratch, and built the Ollie AI assistant end to end, from the streaming UI to its Python side. I also own the hosted MCP server and its OAuth flow.
Open source
I maintain MCP servers for Notion and Replicate Flux and contribute to OPIK's assistant tooling. The MCP packages get around 3.5K npm downloads a week. Counts below are from October 2026.
Owned MCP servers
notion-mcp-server
Typed MCP tools for working with Notion pages and databases. Listed on Smithery and Glama.ai.
replicate-flux-mcp
Generate images with Replicate's Flux models through a small MCP server that's easy to read and adapt.
Ecosystem contributions
Experience
10+ years shipping products in AI tooling, e-commerce, agriculture, logistics, and enterprise software, across four countries. Get in touch
Comet ML
OPIK Product Engineering Own the OPIK frontend; built the TypeScript SDK from scratch and the Ollie AI assistant (UI and Python side); own the hosted MCP server and its OAuth; trace performance on millions of rowsCapgemini Engineering
Enterprise Solutions Led a team of 6 on a crypto trading module inside a US fintech banking product: TypeScript migration, review standards, and test coverageEasyGet
Delivery & Logistics Owned a React Native mobile app end to end in a startup environment: product decisions, analytics, release readiness, and production deploymentRegrow Ag
Agriculture Analytics Made heavy map interfaces 4 to 5x faster and cut data loading time by about 50% with Node, GraphQL, Redis, and PostgreSQLLatest posts
Recent writing on engineering, tools, and open source.
One MCP Tool Call Can Burn Your Whole Token Budget
MCP tool responses flooding the context window? How the Opik MCP server cuts token usage: three detail levels, a token count per answer, no silent cuts.
MCP Tool Design for AI Agents, Not API Endpoints
Long MCP tool lists waste LLM context and confuse AI agents. An MCP server design that swaps endpoint-shaped tools for one execute and one describe tool.
Stop Using Claude Code on Defaults
The Claude Code settings.json changes I use to cut token costs, stop hallucinations and skip permission prompts: model routing, LSP, Gemini delegation and more.
Agentic UX Primitives: The Frontend Patterns Nobody Taught You
Agentic UX patterns for AI agent UIs: streaming with intent, human-in-the-loop gates, reasoning visualization and confidence indicators.
Context Engineering Ate Prompt Engineering
Context engineering vs prompt engineering: why the context you give AI coding tools (CLAUDE.md, MCP servers, skills) matters more than the prompt.
Let's build something
If you're building something ambitious with AI and need a senior full-stack engineer who can ship it end to end, I'd like to hear about it.
Based in Barcelona (CET), working with distributed teams across Europe and the US.