Forward Deployed Engineer and Solutions Architect for AI. I take a real problem, ship a working system fast, and move on to the next one. Less theory, more shipped.
I build agentic tooling: MCP servers, eval harnesses, and the infrastructure that makes LLM agents reliable in production. 13 years in software, the last 3 on applied LLM systems. Based in Krakow. I write about AI engineering for ~10K people on LinkedIn.
| Tool | What it does |
|---|---|
| cc-cache-monitor | Real-time prompt-cache health for Claude Code. Catches the overnight token bleed before it costs you. |
| pixel-mcp | Figma-to-browser visual diffing with escalating validation gates (CV, DINOv2, VLM, human). |
| synthetic-panel | Spins up a panel of synthetic reviewers to pressure-test content and code before it ships. |
| godmode-telegram-mcp | Full Telegram client exposed to AI assistants over MCP. |
| deep-mcp | Recursive Language Models from the paper. Agents that write code to analyze data bigger than their context window. |
Most of these started as a problem I hit in my own workflow, then became something other people could install.
Python · TypeScript · Claude API · LangChain · MCP · RAG · PostgreSQL · Docker · AWS
LinkedIn: linkedin.com/in/dmytro-tolok