Machine Learning Engineer | Ranking & Recommendation | Production ML | Campinas-SP (Brazil)
I build ML systems that run in production and move business metrics. My XGBoost ranking model scores ~4M items a day and lifted revenue by 26%. My LLM systems autonomously resolve 89% of ~60K monthly support tickets. I work across the full lifecycle: feature pipelines, training, champion/challenger A/B testing, and model serving.
-
🎯 Contextual bandit ranking service (FIAP capstone, lead contributor)
Built the model service for an offer recommendation platform. LinUCB, Thompson sampling and a rule-based baseline are served through FastAPI (/rank,/update). Online learning state lives in Redis with per-policy distributed locks. Shadow and active policies support atomic promotion and rollback, with an MLflow registry and Datadog observability. In a paired multi-seed simulation, LinUCB reached 93.8% of oracle reward, against 90.2% for the baseline. -
☁️ Serverless data pipeline on AWS
Daily pipeline provisioned with Terraform: EventBridge, Lambda, S3 (Parquet with Hive partitions), Glue with Polars, and Athena. Deployed through GitHub Actions CI/CD. -
✈️ Flight delay prediction
Classification (XGBoost, LightGBM, Random Forest) with SHAP interpretability, plus KMeans segmentation into operational personas.
-
💼 Data Scientist, Machine Learning at Acerto (2024 to present)
- Designed an XGBoost ranking model (choice probability × projected revenue, with position bias correction) that scores ~4M debts daily. Validated through champion/challenger A/B testing, it lifted revenue 26%.
- Built and own a multi-agent LLM negotiation system (LangGraph, GCP) on WhatsApp, responsible for 2% of total company revenue.
- Built an LLM support system on Zendesk that autonomously resolves 89% of ~60K monthly tickets.
-
📊 Data Scientist at AB InBev (2023 to 2024)
- Built a PySpark/Databricks forecasting architecture for the European beer market at 92% accuracy, validated by A/B test and generating multi-million savings.
-
📈 Data Scientist and Consultant at Kumulus Cloud & Data (2021 to 2023)
- Deployed a computer vision system (YOLO/PyTorch, Azure) for energy tower inspection, saving BRL 100K/month.
- 🤖 Postgraduate in Machine Learning Engineering, FIAP
- ⚙️ Technologist in Industrial Automation, IFSP (Instituto Federal de São Paulo)
- 📄 Co-author, "An open-access WebApp for Inverse Laplace Transform analysis of TD-NMR signals," Magnetic Resonance, 2026. DOI 10.5194/mr-7-39-2026
- ✔️ Microsoft Certified: Azure AI Engineer Associate
- ✔️ Microsoft Certified: Azure Fundamentals, Azure Data Fundamentals, Azure AI Fundamentals