Open code that proves the closed work

Manifesto

A senior engineering portfolio with no public code is a disadvantage — especially when the house philosophy is technological freedom, open source, and zero lock-in. The problem is that our strongest cases (300 million records, +260% performance, on-premise Llama in a clinic) involve proprietary code under NDA.

The ConsoliDados Lab solves this the honest way: instead of exposing a client, we publish derived showcases that demonstrate the technique — RAG over a knowledge base, bottleneck diagnosis, pragmatic local-vs-cloud architecture — with our own data domain, substantive improvements, and clear credit to sources. It is not a cosmetic fork; it is authorial derivation, dockerized, reproducible, and documented in both Portuguese and English.

The premise is simple: if we claim we deliver real code and not slide decks, the Lab is where you verify that without signing an NDA. Each repository is proof the method works — and each one points to the service it sustains in production.

What is coming

• RAG over graphs — Retrieval-Augmented Generation with Neo4j, natural language to Cypher via LLM, local open source models, and accuracy evals. Demonstrates AI agents and automation.
• Performance diagnosis — Node.js apps with classic problems (memory leak, blocked event loop, connection leak, race condition) in broken and fixed branches, with step-by-step profiling using native tooling. The entry funnel for the performance service.
• Additional showcases — extensions with embedding caches, a hybrid local/cloud option via feature flag, and HTTP endpoints for live demos.

The repositories will be published under the ConsoliDados organization on GitHub, with a bilingual README and an architecture diagram. In the meantime, the current open source code is at github.com/JohnnyCarreiro.

Need this in production?

The showcases show the technique; production requires observability, rollback, integration, and knowledge transfer. If you recognize one of these problems in your system, tell us the case — we respond with feasibility within 24 business hours.