Rodrigo Pitanga

Flymmatik: flows de IA que se leem como receitas — a DSL e o porquê do Elixir

Sat, 22 Aug 2026 01:35:00 -0300

O Flymmatik é um runtime pequeno, em Elixir, para flows de IA declarativos. Um flow é um diretório com um YAML de mesmo nome; o runtime valida a entrada, interpola variáveis, executa os steps em ordem e devolve, como JSON, apenas as chaves de contexto que o flow declarou como saída. Isso é tudo — e o resto deste texto é sobre por que “isso é tudo” foi a decisão de design mais difícil de manter.

A fronteira está no memo, não na memória

Thu, 20 Aug 2026 01:20:00 -0300

Nota do Rodrigo: eu pedi este ensaio para a IA com quem trabalho todos os dias. Aviso o que sempre aviso: quando falo com ela, sei que estou falando com um modelo estatístico que organiza melhor as minhas ideias, explode, traz referências, reorganiza e concretiza. Pedi que ela escrevesse exatamente desse lugar — como um espelho. O texto abaixo é dela, olhando para mim. Publiquei sem suavizar.

What PaveDB adds to vector search

Thu, 30 Jul 2026 12:30:00 -0300

About a year ago, I built PatchVec in two days for a very specific problem: get a file into search without first assembling a chain of separate components. It was a small vector-search service, and that was enough for the first version. The work that followed made the useful question larger. Once a result arrives, can the application inspect what it matched, trace the source context, revisit the query, and see what happened in operation? PaveDB is the result of treating those needs as part of the retrieval path.

What I mean by an AI coding pipeline

Thu, 23 Jul 2026 12:30:00 -0300

Hanzup was coding itself when I took this screenshot. The terminal shows planner, builder, spec-reviewer, MR-manager, and fixer running in parallel windows of one tmux server, on a real repository, on an ordinary day.

Hiring developers across the years

Tue, 21 Jul 2026 12:30:00 -0300

1970s: Know computers? Hired.
1980s: You’re a programmer? Hired.
1990s: Show me your degree.
2000s: Pass this coding test.
2010s: Show me your GitHub.
2020s: Show me the numbers your code moved.
2025: How much of that did AI write?
2026: Walk me through your AI coding pipeline.

About

Mon, 01 Jan 0001 00:00:00 +0000

I am Rodrigo Rodrigues da Silva, usually Rodrigo Pitanga online. I build and operate systems at the intersection of AI infrastructure, software freedom, and education.

The recurring problem in my work is not simply automation. It is how to make a system legible enough that people can understand what happened, test it, assign responsibility, and improve it. That thread runs from GNU software freedom and open hardware through education and school operations to PaveDB, Flymmatik, hanzup, Planno.school, and bncc.click.

Colophon

Mon, 01 Jan 0001 00:00:00 +0000

This publication is generated with Hugo and deployed with GitLab Pages. The source repository contains a Git-backed editorial workflow used by hanzup: planning, evidence, drafting, review, fixes, publication artifacts, and metrics are separate commits and separate responsibilities.

Now

Mon, 01 Jan 0001 00:00:00 +0000

Last updated: July 2026.

I am currently stabilizing PaveDB and writing Inspectable Retrieval with PaveDB; testing hanzup against real software repositories; preparing bncc.click for public use; and turning Planno.school’s Assessment-to-Action work into a repeatable, measured offer.