Using AI and soccer match data to build my own commentator
Building the context layer between raw match events and a language model — what the AI got right, where it fell short, and why the pipeline matters more than the prompt.
Read more →Turning complex systems into simple, reliable outcomes.
I design and build backend-heavy systems end to end, choosing monolithic or microservice architectures based on what a given problem actually demands rather than by default.
My day-to-day spans data migrations and ETL pipelines, instrumenting systems with resource and usage metrics to make observability real, digging into error and bug patterns to defuse critical scenarios before they escalate, and reading business context and process flow to turn raw data into decisions people can act on.
That range is how I replaced a six-figure external contract with an in-house platform in three months, and separately turned a manual editorial process into an automated one running directly off structured data.
Lately I’ve been extending that into AI-applied system design — RAG pipelines, LLM integration, semantic search — plus a lot of GCP, BigQuery, and Pub/Sub, in Python, Java, and Go. BSc in Computer Science, Fluminense Federal University (UFF).
Happy to talk more about any of this.
$ skills --list --grouped [AI & Data] RAG pipelines · LLM integration · semantic search · real-time ingestion [Backend] distributed systems · API design · event-driven architecture · data modeling [Infra] container orchestration · CI/CD · IaC · self-hosted tooling [Languages] Python · Java · Go · JavaScript
Featured work
Three elections, three encounters with Brazil's electoral data — and the third time, it was mine to lead
From a junior dev drowning in undocumented CSV files in 2019, to watching a six-figure third-party system fail on election night in 2024, to leading the build in 2026. The story behind the system.
Read more →From GitLab to Gitea: how my homelab CI/CD went from 3.4 GB to 100 MB
The journey of setting up self-hosted CI/CD on a Raspberry Pi 4B — through GitHub Actions ARC, GitLab CE, and landing on Gitea with act_runner, nerdctl, and buildkitd.
Read more →Recent posts
- Behind Intercept Brasil's donation ask: hypotheses on cache, edge, and origin
- Split to double: what I learned bringing two solutions to a hackathon
- I gave Claude 12,732 memories. Here's what it noticed.
- The memory stack under the AI workflow: CLAUDE.md, MemPalace, and a knowledge graph
- Three elections, three encounters with Brazil's electoral data — and the third time, it was mine to lead