christiam27
Remote: Yes
Willing to relocate: Yes (US)
Technologies: Claude Code, Codex, Google Antigravity, Python, FastAPI, Node/Fastify, JavaScript, React, Postgres/Prisma, Supabase, Redis/BullMQ, Playwright, Docker, Linux, Cloudflare, Moodle
Résumé/CV: github.com/manuelleal (CV on request)
Email: [email protected]
I build deployed software by directing a team of AI agents: an architect agent assigns and reviews work, builder agents write specs, code and tests, and independent read-only auditors must approve anything touching money, permissions or keys. Every process error gets logged and numbered. I come from education (teacher trainer, PhD student in Education at UAB), so I can also train non-engineers to use AI well.
- Engrama (engramaingles.com; no-login demo: engramaingles.com/magia): English-learning platform built for institutions. Class game with coins, streaks, store and auctions; adaptive CEFR placement exam plus a Saber 11-style mock (~1,800 items); live Kahoot-style quiz used at a teacher conference in October. FastAPI + Postgres with per-client row-level security and HMAC-signed events; 838 backend and ~974 web tests passing. Self-hosted on Ubuntu, Docker, Coolify, Cloudflare Tunnel and Caddy. The original classroom MVP was used by 97 students. No paying customers yet.
- Helper (private, 136 commits): an unreleased multi-tenant tool I built for SENA instructors on Moodle, which I use with 100+ students: submissions, grading, forum replies, bulk messages, reminders and records. Fastify, BullMQ (15 workers), Redis, Postgres/Prisma, Playwright, React. Claude proposes matches and grades; the instructor approves. It calls Moodle's session AJAX, so one POST returns 147 participants across 3 assignments in 2.7 s versus 2-4 s per page scraped.
- JUACO: artificial-life research run with LLM agents under preregistration, replication and independent audit; 120 agent errors logged; DOI 10.5281/zenodo.23041088. Open method: github.com/manuelleal/juaco-metodo (blind test: the auditor caught 28 of 40 seeded defects).
- Offline AI tutor (my PhD prototype): a small open model on a Linux Chromebook for remote schools; caught 82% of seeded errors in a preregistered blind test, 1.7 s median response on CPU.
- 160+ hours of AI training for teachers.
Looking for: AI automation, AI enablement or agent-built internal tools; contract or full-time.