A self-driven AI product strategist and automation engineer who'd rather build systems than climb ladders. I build the tooling layer that makes AI agents actually useful — MCP servers, resilient browser automation, and the backend services behind them.
~13 hours a day in the shell. I focus on the tooling layer that makes AI agents actually useful — Model Context Protocol servers, browser automation that survives bot detection, and the backend services behind them. I like small, sharp, dependency-light tools that ship and hold up in production.
Roles and highlights.
Build and maintain the crawlers behind app.dimensions.ai — grants & policy data at +500% volume. Shipped scraper tooling with elevated browser access, a cache-based self-healing AI scraper that scouts funding data, and lightweight Jenkins/Jira MCPs plus dozens of agent skills to automate manual flows.
Co-led a 12-person team building a Minecraft-based MMORPG. Wrote Python/Node automation over millions of content files, scrapers monitoring CurseForge & modding communities, custom Java mechanics, and the modpack site (zoompack.dev).
Automated AEM deployments (Bash/Python) to cut release risk, built Jenkins CI/CD for multi-project AEM environments, and tuned AEM Dispatchers / Apache for caching, security and performance.
Built a browser extension to automate clinical data collection and scrapers to extract trial data with rate limiting and smart retries — early automation, before AI went mainstream.
Delivered on-prem apps for a law firm (landing page, vehicle-mileage management) and a restaurant management system.
Led migration of a Shutterstock backend service from Ruby on Rails to Java (JAX-RS) with full feature parity — hardened API endpoints, structured validation, and comprehensive test coverage.
Built and maintained 3 full-stack e-commerce platforms, driving +18% quarterly sales.
Education — B.Sc. Computer Science, „Alexandru Ioan Cuza" University, Iași · 2016–2020
Agent tooling and backend infrastructure — most are open source.
Zero-dependency control panel for llama.cpp that auto-tunes any GGUF for maximum context (MoE expert-offload + quantized KV cache) and serves an OpenAI-compatible endpoint. Runs a 30B MoE coder at 256K context on a 16 GB GPU.
A local, MCP-native cluster where AI agent harnesses delegate work and pick it up as sentries: worker registry, task board, message bus, full-text memory, live dashboard. Dumb pipes, real agents.
Read and continue your local Claude Code chats from your phone over LAN/VPN. Zero-dependency Python server + a polished mobile web UI; live via SSE, full markdown & tables, continues the exact session with claude --resume. No cloud relay.
Agent-compatible skill that pings you on Telegram when you're away from the machine, and takes your reply back into the session. Tuned for 30B+ local models.
Reach for the right tool; keep the dependency count honest.