Senior AI engineer in Lisbon, remote worldwide. 14 years shipping: fintech and e-commerce first, 4+ years deep in web3, now AI-native products and agent tooling. I build agent systems and the loops that keep them honest: bounded specs, isolated worktrees, evals that fail closed. These days that also means automating 3D and animation production (Blender, Unreal Engine, MetaHuman) with agent-driven first passes artists can trust.
This README speaks MCP. Point your agent at https://www.rubenmarcus.dev/api/mcp and let it read the résumé instead of you: get_resume, get_services, check_availability, book_intro. Setup for Claude, ChatGPT, Cursor, Kimi and Codex lives in the connect guide.
- #1 on ECDSA.fail with a 9-role agent campaign across 7+ providers · #1 QEC decoder on Optimization Arena (2,642 EPM)
- Co-author on ECDSA.Fail: Open Autoresearch (arXiv:2609.09582), the paper that came out of it
- Agent runtimes in production: 2.85M+ messages processed, 24,164 users, 16,703 agents deployed
- 34K+ all-time npm downloads across 8 packages
- Tupi · Bertioga channel, 1554: Tupinambá canoes at dawn in real-time 3D (three.js WebGPU), built from prompts with cited historical research. Latest demo drop, prompt on the card.
- ScanRepo · Scan a repo before you clone it: credential stealers, supply-chain traps, fake recruiter scams. 31+ static rules, 6 risk categories, 0 to 100 risk score, nothing cloned or executed.
npx scanrepo owner/repoin CI. - AEO.js · Answer Engine Optimization for JS sites:
llms.txt,ai-index.json, AI-crawler policy analysis. Astro and Next plugins. - AEO Checker · The free scanner on top of AEO.js. 4,569 scans across 2,259 sites.
- CS Brasil · Browser FPS shipped through an agent gauntlet: measured baselines, critics that never grade their own work, A/B frame proof. 2,191 players, 27 countries.
Earlier: Ralph Starter, Autoresearcher, Mirofi.sh. Full 14-year archive →
- This portfolio is agents-welcome. Probably the first.
- Streaming 2.85M messages: the plumbing of a production agent chat
- The AI harness behind the CS Brasil game
- Routing 9 agent roles across 7 providers: the ECDSA.fail harness
Every post ships in EN and PT, and every URL has a .md twin for agents. All posts →
Selectively available for full-time roles and freelance contracts. Fixed scope preferred. I reply within a day or two.
- AI products & agent systems · prototype to product, with evals and observability · 3 to 8 weeks
- AI-native frontend & design engineering · distinctive web product, quality held by the delivery loop · 2 to 6 weeks
- AEO audit & implementation · make your site legible and citable to answer engines · 2 to 4 weeks
ruben@rubenmarcus.dev · contact · or just let your agent call book_intro.