Pietro Grazzioli Golfeto
About Me
I'm a Computer Science student at Unicamp (graduating December 2026). Currently, I work as a Machine Learning Engineer Intern at Nubank AI Core, building Foundation Models and engineering pipelines for the largest digital bank outside Asia. In parallel, I am an Undergraduate Researcher at H.IAAC (Recod.ai), finalizing a publication on autonomous agents for financial modeling and tackling concept drift in non-stationary environments. Previously, I was a Software Developer Intern at SiDi, where I shipped core Speech & NLP models for Samsung's Bixby. My technical background is supported by awards in national and international mathematics Olympiads, ranking in the Top 25% in Brazil at the ICPC Latin America Regionals, and recently becoming a Global Finalist (Top 0.24%) in the NASA Space Apps Challenge 2025.
Projects
- Project A.T.L.A.S. (NASA Space Apps 2025 Global Finalist): Led the A.T.L.A.S. team to the Top 45 globally out of 18,860 submissions. Directed the integration of Generative AI and NASA GIBS data to build a high-performance 3D interactive web experience that personifies NASA's Terra satellite to visualize climate causality.
- Route Orchestrator: Logistics Optimization Suite: Co-developed a high-performance optimization suite for the Heterogeneous Fixed Fleet Vehicle Routing Problem (HFFVRPTW). Built a parallel batch processing engine to execute 236+ instances concurrently and designed a complete statistical analysis framework to rigorously benchmark algorithm performance.
- Autonomous Agents for Financial Modeling: My undergraduate research project at H.IAAC. This work introduces a self-learning framework to adapt to non-stationary economic environments, tackling the challenge of concept drift and sample bias. The goal is to create an adaptive agent that minimizes long-term financial losses, outperforming traditional static credit scoring models.
- Ethical AI for Chikungunya Outcome Prediction: Predicted hospitalization outcomes using models trained with nested cross-validation and resampling techniques to handle severe class imbalance. The project includes a comprehensive fairness analysis that mitigates demographic bias (reducing the fairness gap by ~93%) and achieves equitable performance with a minimal accuracy trade-off.