Open study · Nov 2026 – Present
Brazil's Data Leadership Job Market
Open dataset of data-leadership job postings in Brazil (coordinator × manager), published weekly and automatically from a self-built job-tracking app, with a data contract and confidentiality rules checked before every release.
dbt-on-DuckDB pipeline with tests and CI that needs no database credentials; a site where every number comes from the models, with the SQL that produces it running in the browser; posting lines classified by an LLM chosen by measurement against a hand-reviewed reference set.
Open the study ↗Code ↗
The parts
- Job-tracking appwhere the data is born
- Data contract and confidentialitychecked before every release
- dbt on DuckDBtests and CI with no database credentials
- LLM measured against a referenceclassifies every line of the postings
- Site with the SQL next to itevery number with the query behind it
Personal project · Oct–Nov 2026
Stack Overflow Answerability
Analytics engineering on the public Stack Overflow dataset in BigQuery: specs and acceptance criteria before the SQL, dbt models and tests, a MetricFlow semantic layer, and query-cost guardrails.
LLM-based text labeling, with the labeling model chosen by agreement against a hand-reviewed gold set; calls through the Vercel AI Gateway and direct APIs, with a budget cap per run.
Open the presentation ↗Code ↗
The parts
- BigQuerythe public Stack Overflow dataset
- dbt models and testsspecs and acceptance criteria before the SQL
- MetricFlowsemantic layer
- Vercel AI Gatewaycalls with a budget cap per run
- LLM + gold setlabeling measured by agreement
A method turned into a product · since Sep 2025
Matriz de Senioridade
A method for building competency matrices from each team's context, for assessment, feedback, merit, promotion and development plans on explicit criteria. It started in the data team I led, was adapted by other leaders and by HR, and became a SaaS built with AI agents.
Open the Matriz ↗
The parts of the method
- The team's contextwhere every matrix starts
- Competencies and levelssteps on the Dreyfus model
- Assessment and feedbackthe conversation on top of the criteria
- Merit and promotiondecided on the same criteria
- Development planeach person's next step
CRM and martech · Sep–Oct 2026
WhatsApp CRM and Campaigns
Customer-base analysis for both businesses: ingestion from their sales systems, normalization, fuzzy deduplication of customer records, and repeat-purchase and cross-sell analysis.
WhatsApp Cloud API integration (webhook with signature verification) and an inbox with a campaign funnel in production, used by the business, with send guards and an A/B/C test; built with AI agents.
A brewery and a bar in Porto Alegre · private code
The parts
- Ingestion and normalizationfrom both businesses' sales systems
- Fuzzy deduplicationof customer records
- Inbox + funnelcampaigns in production, used by the business
- A/B/C testwith send guards
- WhatsApp Cloud APIwebhook with signature verification
