Work
Public repositories by Guilherme Cavalcanti under ARBOCK LABS. For each project we show what it is, what was tested, and what was not. Test counts come from runs recorded on 2026-10-19 unless stated otherwise.
What it is: a full-stack Next.js application for client records and commercial proposals, with authentication, per-user persistence, line-item totals and server-generated PDFs. Tested: 24/24 unit tests passed; typecheck clean. The "Web validation" CI workflow passed on the current commit. Not done: no hosted demo; Docker image not verified.
What it is: a chess analysis environment. A React interface talks to a native chess engine only through a WebSocket controller and a UCI adapter; an Android build runs the engine as a native process. Tested: 8/8 parser and session tests passed (mocked engine output); typecheck clean. Not done: no CI; engine smoke tests not run; no strength or performance benchmark.
→ SHERLOCK
What it is: synthetic form → rules-based triage → editable suggestion → human approval → log in a private 24-hour session.
Tested: 6/6 local tests passed; remote CI passed; main flow checked at the public URL.
Limitations: original source unavailable; Portuguese interface; no real delivery, CRM or generative AI.
→ Case study · Open demo
An offline chess line trainer: the app shows a line, you repeat it, and a wrong move resets the drill. 21/21 tests passed.
How these projects were built, through human-directed AI-assisted development, is documented separately.
Describe what you have and what is not working.