Lucca De Carvalho Matuck
Founder and Engineer
Experiências
- 06/2026 até 11/2026 (5 meses)
Founder and Engineer
Em Camadas 3D
Built 3 Claude Code plugins (21 skills, 6 sub-agents, among them a planner, a fresh-eyes reviewer, an analyst and a strategist) for the business, plus my own MCP server for the Mercado Livre seller API. The human-in-the-loop content pipeline turns raw footage into scheduled posts on four networks; every external publish requires an explicit human approval of a preview, per network. ~203k organic views. Open-sourced a local timelapse studio with classical computer vision (linear-light colour correction, ORB+ECC stabilisation): 95 tests, CI.
- 12/2024 até 11/2026 (1 ano e 11 meses)
Applied AI and Systems Developer
Freelance, restaurant client
Sole engineer on a restaurant operations system (orders, kitchen queue, stock, purchasing, point of sale) with an LLM WhatsApp ordering assistant, built and in operation at the client. React/TypeScript, PostgreSQL/Supabase, 105 migrations, built with Claude Code. The model proposes, the code disposes: one strict JSON schema response per turn, re-validated against the current catalogue; prices come from the database; the only write is one order-creation function under a least-privilege Postgres role. Delivery zone is decided in code (geocoding, point-in-polygon); failures hand off to staff. Row Level Security on all tables, with 11 roles modeled on the restaurant’s staff positions; immutable payment and stock ledgers; 54 test files, including a 26-scenario end-to-end run of the assistant against a real local database with the LLM, WhatsApp gateway and geocoder faked.
- 12/2021 até 02/2026 (4 anos e 2 meses)
Doctoral Researcher (Machine Learning for Sensor Data)
University of Aveiro/i3N
Built regression models that recover internal battery temperature and pressure variation from five external signals, on held-out cycles of one instrumented cell per chemistry (Li-ion, Na-ion): 0.075°C and 0.048bar RMSE, R2 up to 0.996. Benchmarked five architectures (ridge, random forest, SVR, LSTM, TCN) in TensorFlow and scikit-learn under Bayesian optimisation, with cycle-level splits preventing leakage. Wrote the full signal chain, from raw interferometric spectrum to physical quantity, then ported it to Python and cross-checked it against the original MATLAB on 7,902 recorded spectra: within 0.17pm worst case.
Formação
Engineering Physics
University of Aveiro
2021 até 2026 · Doutorado - Completo
Mechanical Engineering
University of Aveiro
2016 até 2021 · Graduação - Completo
Idiomas
- Português Nativo
- Inglês Proficiente
- Espanhol Conversacional
Habilidades
Perfil completo, com vídeo e mais detalhes: recrutei.me/vfj3qqe4