Inspectable AI infrastructure
Retrieval, provenance, replay, evaluation, and systems whose retrieval behavior can be examined.
Signal / Action
Inspectable systems from signal to action.
I work across AI infrastructure, software freedom, and education: retrieval you can inspect, agent workflows with bounded authority, and school operations that turn signals into action.
The work
Three lanes, one operating thesis
Governed agent delivery
Multi-agent software work with narrow roles, bounded permissions, Git-backed state, and human gates.
Education systems
Turning assessment, planning, institutional memory, and school data into governed action.
Latest
Writing
Flymmatik: flows de IA que se leem como receitas — a DSL e o porquê do Elixir
O desenho do Flymmatik: uma superfície mnemônica para flows de IA declarativos, os princípios que a seguram, e por que o runtime é Elixir/OTP.
ai coding pipelinedsl designelixir
A fronteira está no memo, não na memória
Um ensaio escrito pelo modelo que trabalha comigo, sobre três semanas em que escrevemos mais documento que código — e sobre por que foi exatamente isso que fez tudo funcionar.
ai coding pipelinedocumentationengineering practice
What PaveDB adds to vector search
PaveDB turns file ingestion, retrieval provenance, query records, and operations into one inspectable path for retrieval applications.
retrievalvector searchrag
What I mean by an AI coding pipeline
How a role-based coding pipeline separates implementation, review, and remote delivery authority.
agent systemsai engineeringsoftware delivery
Hiring developers across the years
How the signals that reveal a developer's value have shifted from credentials to visible engineering work.
software developmentai engineeringcareers
Current systems
Projects
active
PaveDB
Inspectable retrieval database
A small retrieval substrate built around provenance, query history, replay, local operation, and explicit boundaries.
draft
Inspectable Retrieval with PaveDB
Technical book
A book about building RAG and semantic-search systems that teams can deploy, operate, observe, evaluate, and trust.
coming soon
Flymmatik
Declarative execution layer
A small runtime for structured intelligence services, keeping deterministic steps, inference, resources, and channels explicit.
Follow the work
Essays and field notes, as they’re published.
Read new posts through RSS, or follow the shorter notes on LinkedIn.