Building5 updates
RegWatch
Google Alerts for the Diário Oficial da União (DOU)
A monitoring service that watches Brazil's federal gazette (the DOU) and alerts professional-services firms when their clients or topics are mentioned.
View build log →Building Publicly
Progress updates from the projects I'm building in the open. Finished projects can be found under Work.
Building5 updates
ReplayGate
Cross-turn regression testing for multi-turn, channel-native AI agents
A harness that records a real agent conversation — its LLM and tool calls — into a deterministic fixture, then replays it to assert cross-turn invariants like "never books before the user confirms": the whole-conversation bugs that per-turn tests miss, on the channel the agent actually runs on.
Latest
The LLM judge that can't fail your build: an offline evaluator, advisory by defaultUpdate 058 min
Building5 updates
Turmarium
Multi-tenant school-management SaaS with tenant isolation enforced in the database
A B2B SaaS where a school registers its people and academic structure (disciplines, syllabi, terms, offerings/turmas, sessions, grades, attendance). Built English-first with pt-BR first-class as a Django modular monolith: the academic core is the stable spine, and every future capability (gamification, AI, exams) plugs in as a module through published contracts and domain events, never by importing the core. Tenant isolation is enforced at the database with Postgres row-level security.
Latest
White on white: making dark mode a token decision, and the palette that failed its own contrast testUpdate 059 min
Building3 updates
stealthbench
A reproducible benchmark that puts numbers on browser-automation stealth
Scores browser-automation stealth setups (configs) against open, self-hosted bot detectors on your own localhost, then commits a versioned results file and a regenerable report. It measures detectability only: no crawler, no login, no target site, and every detector runs offline.
View build log →Building1 update
supskill
A sprint conductor for Claude Code
Drives one sprint from a markdown backlog to a merged branch: a fresh context boundary at every stage, three human gates, and all authoritative state on disk so the conductor survives its own context death. A pure-Python CLI is the only thing allowed to mutate state and refuses any transition whose preconditions aren't met; the conductor skill runs the gates in the one place a human can actually answer.
Latest
Building on a moving runtime: I re-ran the finding my sprint conductor is built onUpdate 018 min
Follow the build
Build-in-public updates and new posts, delivered as a digest. Double opt-in · no spam · unsubscribe anytime · handled by Buttondown.