I design and build production-grade distributed backends, end-to-end machine learning pipelines, and
accessible full-stack web applications. My work spans quantitative finance, energy
trading, and global eCommerce automation across three countries. IEEE member, academic
researcher, and active FOSS contributor.
Based in São Paulo, Brazil.
I'm a full-stack software engineer based in São Paulo, Brazil, with
hands-on experience across the entire product lifecycle: architecting distributed systems
and optimizing database schemas at scale, training and deploying production machine learning models,
designing REST and GraphQL APIs, and crafting polished, accessible user interfaces.
My professional track record spans three distinct domains: at Giant Steps Capital,
Latin America's largest quantitative asset manager, I built post-trade systems and
data engineering pipelines handling billions of dollars in daily transactions; at
Libra Energia, I led the containerization of the full internal application
stack and engineered data science tools for hydric forecasting; and at
FlxPoint, a US-based multi-tenant eCommerce SaaS platform, I currently
build custom machine learning-powered catalog solutions and REST APIs for enterprise clients. Each
environment has demanded a different mix of technical depth, system reliability, and
cross-functional collaboration.
Open-source software has shaped how I work and think as an engineer. I author and
maintain several FOSS libraries, spanning Python date utilities, Rust CLI
tooling, pre-commit hooks, and GitHub Actions for static analysis and dead-code detection.
I also contribute to widely-used projects including Ruff, Django REST Framework, OpenBB,
and more. Building in the open is one of the clearest ways I can give back to the
developer community that shaped me.
I am currently completing a Master's degree in Electrical Engineering
and Computing at Mackenzie Presbyterian University, with research centered on multimodal
drone detection through fusion of deep learning-based computer vision, radar signal
processing, and radio frequency analysis. My undergraduate thesis, a quantitative
evaluation of image preprocessing's effect on dermatological AI accuracy using the
HAM10000 dataset, is under review at IEEE Transactions on Image Processing.
I am an active member of IEEE and the
Brazilian Computer Society (SBC).
High-throughput backend systems Machine learning UI/UX prototyping & implementation FOSS contributor Cloud & DevOps AI Research
› Designed, developed, and maintained custom eCommerce solutions tailored to client requirements on a multi-tenant SaaS platform.
› Engineered full-stack solutions using Vue.js + TypeScript for front-end, Java (Spring Boot, jOOQ) for back-end, AWS for cloud automation, and Flyway/PostgreSQL for data management.
› Developed custom machine learning models to streamline customer product categorization and publishing workflows.
› Built an in-app chatbot using RAG and LLM APIs, significantly reducing customer support and onboarding time.
› Delivered customer-facing REST APIs for advanced programmatic platform usage.
› Built post-trade systems for Latin America's largest quantitative asset manager, including web tooling, REST APIs, data conciliation pipelines, and automated billing routines.
› Engineered worldwide stock exchange messaging systems and critical data engineering pipelines handling billions in transactions.
› Developed high-performance eCommerce marketplaces using Magento, Akeneo PIM, and ScandiPWA.
› Engineered comprehensive eCommerce solutions from back-end infrastructure to React front-end interfaces.
› Focused on scalable architecture and seamless user experience for evolving client requirements.
PHPMagentoReactScandiPWAAkeneo
// education
Academic Background
Degrees
2026 – 2027In progress
Master's Degree
Electrical Engineering and Computing
Mackenzie Presbyterian University · School of Computing and Informatics
Research focus: Multimodal Drone Detection (computer vision, radar, RF signals)
2022 – 2024
Bachelor's Degree
Computer & Information Systems
Mackenzie Presbyterian University · School of Computing and Informatics
Undergraduate thesis published in IEEE Transactions on Image Processing (under review)
2022 – 2024
Associate's Degree
Computer Systems Analysis and Development
Mackenzie Presbyterian University · School of Computing and Informatics
Certifications & Honors
AWS Academy Machine Learning Foundations
Amazon Web Services · 2024
AWS Academy Cloud Foundations
Amazon Web Services · 2024
Red Hat System Administration I & II
Red Hat · 2024
Test of English as a Foreign Language (TOEFL)
ETS · 2018
Certificate of Proficiency in English (CPE)
Cambridge · 2016
// research
Academic Research
IEEE Transactions on Image Processing
Under Review2024
Quantitative Analysis of the Impact of Image Preprocessing on the
Accuracy of Computer Vision Models Trained for the Identification
of Cutaneous Dermatological Diseases
Gabriel Mitelman Tkacz
& Gustavo Scalabrini Sampaio
· Funded by CAPES Foundation
A systematic quantitative study evaluating the impact of image preprocessing pipelines
on CNNs trained to classify dermatological skin diseases using the HAM10000 dataset.
Introduced the weighted alpha (αw) metric to integrate computational
cost into accuracy gain evaluation. Tested 64 distinct preprocessing combinations —
revealing that simpler pipelines often yield superior cost-adjusted performance, and
that baseline model accuracy fundamentally modulates preprocessing effectiveness
(ceiling effect). Top pipeline achieved +9.5% accuracy gain with 13× computational
efficiency advantage.
Key Findings
Best single preprocessing: Equalization alone (+8% α, αw = 10.97)
Best multi-step: E → CS → N achieving max gain with αw = 13.17
~50% of all tested pipelines actively degraded model performance
Denoising operations impose 6.5× baseline training overhead
Ceiling effect compresses preprocessing benefits 24× near 100% accuracy baselines
Gabriel Mitelman Tkacz
· Mackenzie Presbyterian University
· Funded by CAPES Foundation
Developing a multimodal drone detection and classification system that fuses
heterogeneous sensor data streams to achieve robust detection across varied
environments and drone classes. The system combines deep learning-based computer
vision with radio frequency (RF) spectrum analysis and radar signal processing,
exploring early, late, and hybrid fusion strategies to maximize detection accuracy
and reduce false positive rates.
Drone DetectionComputer VisionRadar Signal ProcessingRF/SDR AnalysisMultimodal FusionDeep LearningSecurity Applications
Research Roles
Engineering Infrastructure Coordinator
Student Academic League of AI & Data Science · Mackenzie Presbyterian University
Orchestrated DevOps pipelines for machine learning model development and deployment; implemented synthetic data generation and continuous performance monitoring to detect model drift.
Research Assistant
Centre of Regulation and Democracy · Insper Research Institute
Applied machine learning, NLP, and web scraping to bridge computer science and legal research, streamlining juridical research processes.
SOLID PrinciplesGRASPClean ArchitectureREST API DesignGraphQLScrumKanbanTDDCI/CDDatabase Optimization
// open source
FOSS Contributions
I'm a strong believer in open-source software. I author and maintain libraries used by
developers worldwide, and regularly contribute bug fixes, features, and documentation
to projects I rely on every day.