Gilberto De Melo Junior

Applied Data Scientist with 4+ years of experience delivering business analytics, predictive machine learning, and decision-support solutions across workforce, commercial, competitor, and customer data. Strong in Python, SQL, Databricks, distributed PySpark, Power BI semantic modeling, API integration, and data validation, with practical experience prototyping LLM agents and building parallel GPU software in CUDA and Apple Metal.

Senior Data Scientist - Thermo Fisher Scientific - São Paulo, SP

(2026-06)

• Reduced Latin American commercial-headcount processing from approximately two hours to seconds by integrating Workday-origin data with leadership and exception tables in Python and Databricks; classified more than 21,000 records, identified over 7,000 active employees, and persisted results in Delta for Power BI.
• Designed the Power BI semantic model, relationships, DAX measures, and daily dashboard used by Latin American HR teams; replaced more than 12 manual reports, saved approximately two hours per report cycle, and automated semantic-model changes through reusable Python, Power BI API, and OAuth SSO workflows.
• Improved a competitor-identification ML pipeline using Logcomex importation data through data cleaning, standardization, validation, and hyperparameter tuning, increasing accuracy, recall, and F1-score while reducing manual analysis effort for Market Intelligence.
• Prototyped LLM agents with ChatGPT and the OpenAI API to support internal research, analysis, and workflow automation.

Data Scientist - Natural Computation Laboratory, Mackenzie University - São Paulo, SP

(2024-08 - 2026-04)

• Developed and evaluated GraphSAGE, GCN, GAT, and MagNet models in PyTorch Geometric for directed graph classification, achieving 100% recall on 2,001 previously unseen positive graphs at 58.7% precision.
• Wrote all CUDA and Apple Metal kernels and parallelization logic for MaRGS, a GPU-accelerated graph simulation framework, reducing runtime by 90% from more than four hours to approximately 20 minutes.

Data Analyst - Petz - São Paulo, SP

(2022-11 - 2024-04)

• Analyzed behavior across more than 10 million customers using Apriori, FP-Growth, segmentation, and market-basket analysis, generating product-affinity insights that informed customer recommendations during development of the Petz loyalty program.
• Reduced large-scale analytics processing time by 50% through parallel task execution and distributed PySpark workloads in Databricks.
• Automated AWS QuickSight dashboards, replacing recurring Excel and PowerPoint workflows and enabling self-service analytics.

Bachelor of Computer Science - Computer Science - Universidade Presbiteriana Mackenzie (2023-01 - 2026-12)