At Nubank, I work underneath the dashboards, owning data engineering for two operations domains across modeling, data contracts, serving, and observability. I replaced a fragile join with deterministic correlation, bringing the null rate on key fields down from 7.76% to about 0.16%.
At Sinch, I designed a statistical anomaly detection framework that reduced global data platform costs by 40%, and migrated telecom billing pipelines processing 2 billion monthly events from Dataflow to dbt Cloud. At will bank, I scaled invoice processing from 500 to 16,000 emissions per run with asynchronous Python and Athena, without additional infrastructure cost.
I'm also building on the applied AI side: I presented a distributed multi-agent architecture using A2A and MCP at TDC 2025, and developed a clinical RAG assistant evaluated with an automated RAGAS harness. My thesis architecture and evaluation set are open source; the ablation study found the embedding model mattered more than the generator.
