Felipe G. G. Marcelino - Data Scientist

Download Curriculum Here

Summary

Data Scientist with over 6 years of experience, specializing in machine learning, statistical modeling, big data, data analysis, and soware engineering. Proven ability to communicate effectively with teams and stakeholders, transforming business problems into AI solutions using data-driven insights and visualizations. Extensive experience in finance (fraud detection), industry 4.0 (steel quality prediction), healthcare (glaucoma detection via computer vision), resource allocation (reinforcement learning), retail (demand prediction and pricing), and energy (price optimization). Proficient in Python and Scala, with strong soware engineering skills. Adaptable to different environments, excellent communicator, and team player. Holds a Master’s degree in Artificial Intelligence and a Bachelor’s degree in Computer Science.

Hard Skills

  • machine learning
  • python
  • statistical modelling
  • pandas
  • scikit-learn
  • pytorch
  • research
  • apache spark
  • big data
  • data visualization
  • data analysis
  • data cleaning
  • etl
  • model pipeline

Soft Skills

  • communication
  • teamwork
  • problem-solving
  • adaptability/flexibility
  • storytelling
  • critical thinking
  • business understanding
  • proactivity

Work Experiences

  • Data Scientist - Nubank (Dec 2021 - Jul 2024)

    • Develop project on KYC (Know you customer) that involves fraud identity, where the fraudsters try to be another person and obtain any credit product from Nubank. Using automatic retraining procedure and including new features to the model we had a improvement of 20% on precision and 15% on recall making the operation to catch fraudster less costly and less friction to enter on Nubank and legitimated customers to acquire products.
    • I develop model and maintain to catch mule account models, using transactions from different types (PIX, Debit, Withdrawal and, etc.). A mule account involves using an account to hide the origin of the fraudulent money from any kind of criminal activity. Using graph features, the model improved 27% on the precision and, 31% on recall, making the operation cost to catch fraudster decreasing almost 37%.
    • Besides that, I participate on D&I task force to support minority inside Nubank.

  • Data Scientist - Kunumi (Dec 2020 - Dec 2021)

    • Demand Prediction: The goal was to prevent products from going out-of-stock, ensuring they were always available for customers to purchase. Boosting models were used on time series data to determine, within a range, the optimal number of units (SKU) for a specific point of sale across the entire country.
    • Price Elasticity: Using a regression model, the idea was to select a specific price that would yield the highest possible profit for the company through the sale of units of different SKUs. The price was chosen based on the line drawn by the regression. The resulted model increase the revenue in 18%
    • Resource Allocation: Using reinforcement learning methods, the idea was to determine the best number and specific boats to be used in the docking of cargo ships. Fuel consumption varied for each boat due to different power levels. Additionally, as different cargo ships arrived at different ports, it was also necessary to use parallel localization and allocation. The resulted model decreases the cost to allocate the correct resource, plus the usage of fuel, decreasing on total of 4%.

  • Data Scientist/Research Assistant - Federal University of Minas Gerais (Sep 2018 - Dec 2020)

    • APERAM Project:
      • Develop and generate explanations for Machine Learning models in the steel manufacturing process.
      • Pre-process and prepare data for ML tasks.
      • Train ML models on steel manufacturing data to predict defects in steel sheets.
      • Use explainable ML techniques to clarify model decisions, providing insights to engineering experts on how to improve steel recipes.
      • The result of the project generates a paper cited on publication section
    • PSR Project:
      • Develop Machine Learning models for the dispatch of electric power production from multiple sources.
      • Pre-process and prepare data for ML tasks.
      • Train ML models using climate and structural data from energy production sources, such as hydrothermal plants, using reinforcement learning techniques.
      • Create baselines for comparisons.
      • Analyze results and consistency in collaboration with the partner.

  • Data Scientist Intern - Tecsinapse (Jun 2017 - Aug 2018)

    • Nanoparticles:
      • The overall goal of this project is to identify and develop a computational model, using neural networks, capable of predicting and simulating the green synthesis of metal nanoparticles based on real data generated by experiments conducted through the partnership.
    • NLP Sentiment Analyses:
      • The overall goal of this project is to use natural language processing models, bi-LSTM and GRU, for sentiment analysis. The textual data provided by salespeople at a car dealership is processed by the model, and through sentiment analysis, the probability of a customer returning to finalize a purchase is determined.

  • IT Project Manager - Informatica Junior (Jun 2016 - Aug 2017)

    • It is volunteer project with real professional market experience
    • Responsible for managing the development projects of evolutionary features for a web platform application and assisting in development. Application of the Scrum methodology.
    • Responsible for supporting the execution of the directorate’s processes, such as creating commercial proposals, researching the technical competencies of company members, and monitoring projects. Execution of projects using JavaScript (Node.js, Express), WordPress, and the MVC pattern.
    • Responsible for managing the commercial aspects of the organization, aiming to increase the efficiency of project closures. Weekly meetings with clients to finalize projects and provide progress reports.

  • Research Assistant - Federal University of Minas Gerais (Nov 2015 - Nov 2016)

    • Which languages are used together
    • Determine if the developer’s profile changes during project transitions
    • Identify which profile change has the greatest impact on productivity: from backend to frontend or vice versa
    • Identify the most used tools by developers of a certain profile (OS, IDE, etc.)
    • Determine the times of highest productivity for developers

Education

  • Master Degree - Artificial Intelligence - Federal University of Minas Gerais (2019 - 2023)
  • Bachelor - Computer Science - Federal University of Minas Gerais (2013 - 2018)

Publications

  • Predicting Heating Sliver in Duplex Stainless Steels Manufacturing through Rashomon Sets. IJCNN 2021 (Gianlucca L. Zuin, Felipe Marcelino and, Etc.)

Misc

  • English: Advanced
Links to this page