Eduardo M. R. Oliveira

Senior Software Engineer PhD in Computer Science

About me
Eduardo Mucelli

About me

I'm a Senior Software Engineer at Synthesia with a PhD in Computer Science from École Polytechnique. I build backend systems and AI infrastructure, drawing on research in large-scale data analysis and modeling. I collaborate with research teams to evaluate ideas and bring them into production.

I also build and maintain independent products and open-source software, including Docking, a desktop dock for Linux, BSD, and Windows, and CEP Aberto, an open-data platform for Brazilian postal codes with 50K+ users. I've been a Linux and open-source enthusiast since 2003.

Education

École Polytechnique

PhD in Computer Science

Researched how human mobility affects mobile network traffic, planning, and deployment using large-scale datasets covering millions of users. Used Python's multiprocessing and threading tools and R's multicore libraries to analyze the data.

Published five conference papers and two journal articles as first author. The research also resulted in a doctoral thesis and a synthetic mobile traffic generator.

Federal University of Minas Gerais

Master's degree in Computer Science

Studied how network centrality can guide routing in wireless sensor networks. Proposed topological metrics, distributed algorithms for computing them, and tree-based routing algorithms that use those metrics.

Published one conference paper and one journal article. The research is documented in my master's thesis.

Pontifical Catholic University of Minas Gerais

Bachelor's degree in Computer Science

Developed a fire-detection application for wireless sensor networks using real weather measurements. Proposed a method for combining and encoding sensor data.

Published one conference paper.

Professional experience

Synthesia

Senior Software Engineer

  • Improved the reliability, scalability, and observability of Synthesia's AI video generation platform across task scheduling, rate limiting, retries, failure classification, diagnostics, and production incident follow-ups.
  • Migrated core video generation tasks into dedicated services and workers for video preprocessing, rendering preparation, thumbnail and GIF generation, and diagnostics. Ran GPU rendering experiments.
  • Led scaling experiments that reduced compute usage by about 56% with minimal user impact and indicated potential infrastructure savings of about $800K per year.

Canonical

Software Engineer II

  • Implemented kernel module signature verification and failure reporting in Canonical Livepatch’s Go client, with corresponding server-side handling of patch status.
  • Developed and maintained scheduled Python data workflows using Argo on Kubernetes, including PostgreSQL view management, data snapshots, and operational diagnostics.

Canonical

Software Engineer

  • Developed Go services connecting Salesforce with Ubuntu Advantage subscription provisioning and renewals. Optimized PostgreSQL queries and external API calls.
  • Introduced acceptance tests written in Gherkin and run with Godog, covering product listing, purchases, trials, upgrades, and downgrades. Integrated the suite into CI.
  • Refactored Python ETL tooling that synchronized Salesforce and other business data into PostgreSQL, introducing typed interfaces, strict static checking, retries, and operational monitoring.
  • Implemented request tracing across HTTP services and asynchronous jobs, and expanded Prometheus-based monitoring.

BlaBlaCar

Software Engineer

Developed Java services for BlaBlaCar's Trip Search team, covering search, routing, and matching.

  • Developed the meeting-point service, combining PostgreSQL/PostGIS spatial search, machine-learning ranking, Redis integration, and R-tree spatial indexing.
  • Built a geospatial crawler and ingestion tools for a meeting-point dataset containing over one million place records across 14 countries.
  • Benchmarked XGBoost inference libraries for Java and built a reusable prediction library for Java services.
  • Built a reusable Java geospatial library for country and timezone resolution, polyline handling, and geohashing.
  • Improved trip-search indexing and feature ingestion with streamed database reads, pipelined Redis writes, and support for driver-selected routes.
  • Extended routing services with cached Google Directions queries, alternative routes from Google and OSRM, and parallel requests with retries and timeouts.

Orange

Postdoctoral Researcher

Dito Internet

Software Engineer

Developed TIM Beta for Telecom Italia Mobile as part of a Ruby on Rails team.

Telecom Italia Future Centre

Research Intern

Developed Future of Enterprises, a Python application for real-time collaboration among employees.

Task Internet

Software Engineer

Led development of a social network in Ruby on Rails.

Projects

Elevationat

Free API for worldwide elevation lookups from latitude and longitude, using open elevation datasets. Built with Python and FastAPI, with a resumable ingestion pipeline that prepares compressed raster tiles and a spatial index for efficient lookups.

Python
FastAPI
Rasterio/GDAL
SQLite RTree
Docker

Brasil em Reais

Scroll-driven visualization of wealth in Brazil, combining open data on wealth distribution, politicians' declared assets, political salaries and benefits, and estimated fortunes. Includes source-backed comparisons, an interactive population curve, and guided navigation. Runs entirely in the browser, with no backend.

JavaScript
SVG
HTML/CSS
Open data

Docking

Desktop dock and application launcher for Linux, BSD, and Windows, written in Python with GTK 3 and Cairo. Provides window previews, themes, drag-and-drop organization, and 60+ applets, with packaging and CI support for multiple Linux distributions.

Python
GTK 3
Cairo
Linux
BSD
Windows
GitHub Actions

Rosto.io

Facial analysis SaaS that detects faces and landmarks, estimates attributes such as age, gender, emotion, and ethnicity, and detects glasses and smiles. Also provides hair segmentation, head pose estimation, face bounding boxes, and image quality measurements for exposure and blur. Predictions use models built with Keras and TensorFlow. See the API documentation for supported features and request examples.

Python
Keras
TensorFlow
Ruby on Rails
Docker
Nginx

Eazy.bike

Bike-sharing route planner with a catalog of 779 cities across 53 countries. Uses the latest reported bike and dock availability to recommend pickup and return stations. The Android app reroutes riders if their destination station fills up during a ride. City-specific domains, including paris.eazy.bike and london.eazy.bike, provide local maps and route planning.

Ruby on Rails
Python
Java (Android)
MySQL
jQuery
Apache

CEP Aberto (Open postal code)

Collaborative open-data platform for Brazilian postal codes (CEPs), covering about one million codes and serving 50K+ users. Provides a developer API and community tools for adding and correcting postal-code data.

Ruby on Rails
Python
MySQL
jQuery
Apache

Proconfie

Application helping consumers compare companies using complaint data. Received an honorable mention in the Brazilian Ministry of Justice's Open Data Applications Contest.

Ruby on Rails
R
MySQL
jQuery
Apache

Benchmark XGBoost Java

JMH benchmark suite comparing prediction latency across XGBoost libraries for Java, including XGBoost4J, JPMML, XGBoost Predictor, and rJava. Evaluates linear and tree-based models across multiple use cases and latency percentiles. The results are discussed in Fast Machine Learning Predictions.

Java

Selected publications and technical writing

See Google Scholar for my full publication list.

Fast Machine Learning Predictions
BlaBlaCar Tech Blog
Eduardo Mucelli Rezende Oliveira
Machine Learning for Predicting QoE of Video Streaming in Mobile Networks
IEEE ICC Communications QoS, Reliability, and Modeling Symposium
Yu-Ting Lin, Eduardo Mucelli Rezende Oliveira, Sana Ben Jemaa, Salah Eddine Elayoubi
Mobile Data Traffic Modeling: Revealing Temporal Facets
Elsevier Computer Networks
Eduardo Mucelli Rezende Oliveira, Aline Carneiro Viana, K. P. Naveen, Carlos Sarraute
On the regularity of human mobility
Elsevier Pervasive and Mobile Computing
Eduardo Mucelli Rezende Oliveira, Aline Carneiro Viana, Carlos Sarraute, Jorge Brea, Ignacio Alvarez-Hamelin
Analysis and Modeling of Mobile Data Traffic in Mexico City
NetMob
Eduardo Mucelli Rezende Oliveira, Aline Carneiro Viana, Naveen K. P., Carlos Sarraute
Measurement-driven mobile data traffic modeling in a large metropolitan area
IEEE Pervasive Computing and Communications (Percom)
Eduardo Mucelli Rezende Oliveira, Aline Carneiro Viana, Naveen K. P., Carlos Sarraute
From Routine to Network Deployment for Data Offloading in Metropolitan Areas
IEEE Sensor and Ad Hoc Communications and Networks (SECON)
Eduardo Mucelli Rezende Oliveira, Aline Carneiro Viana

Languages

Portuguese
Native speaker
English
Fluent
French
Fluent
Italian
Advanced

Awards

Honorable mention
Brazilian Ministry of Justice

Lead developer of Proconfie, which received an honorable mention in the Open Data Applications Contest.

Silver Medal
Pontifical Catholic University of Minas Gerais

Ranked second in the bachelor's degree program.

Honorable mention
Pontifical Catholic University of Minas Gerais

For the work “Game Theory for Decision Making in Wireless Sensor Networks,” presented at the 15th Science Seminar.

Contact me