João M.

Deep Learning Research Engineer

Netherlands

Trusted member since 2025

10 years of experience

He specializes in building advanced models, including large language models (LLMs), capable of code refactoring, bug detection, and continual learning. João works extensively with PyTorch and deploys models on cloud platforms and high-performance computing systems.

Prior to ASML, he led research teams at GAIPS Lab, published in leading AI conferences, and secured competitive grants from the U.S. Air Force and FCT. He also taught AI courses, earning a Teaching Excellence Award for his contributions to education.

João’s key projects include advancing continual learning techniques, enabling AI to acquire new knowledge without forgetting previous tasks, and applying reinforcement learning to train models more efficiently with less data. He is passionate about making AI systems more effective, practical, and continually improving.

Main expertise

Python10 years

Machine Learning10 years

Data Science10 years

Scikit-learn10 years

12+

Experience7

AI Researcher

Utrecht University•

Artificial Intelligence (AI)

Jan 2026 · 9m

Developing a production-oriented AI system for personalised patient interventions in dementia care, combining reinforcement learning with large language models. Designed and built ToneRL, a hybrid architecture where a lightweight RL agent learns to control an LLM’s output in real time, adapting communication style to individual patients based on clinical feedback, without retraining or modifying the underlying model. Collaborating directly with clinicians and linguists to ensure outputs meet clinical communication standards. The architecture is designed for scalable deployment: one frozen base model serves all patients, with per-patient adaptation handled by lightweight policy instances requiring minimal compute.

MongoDB

Docker

Project Management

Budget Management

Python

45+

Deep Learning Research Engineer

ASML•

Artificial Intelligence (AI)

Nov 2024 - Jan 2026 · 1y 2m

Led a research team on the "LLMs for Software Engineering" project, focusing on technical debt reduction, bug detection, and documentation analysis using Large Language Models.
Designed, implemented, trained, tested, and deployed LLMs for automatic code refactoring and bug detection.
Deployed models to cloud production environments and HPC distributed computing clusters.
Monitored the continual performance of deployed models using tools such as MLFlow, Sacred, and Weights & Biases.
Connected the company’s research department with academic partners at TU/e.

Docker

Java

Flask

Python

C++

53+

Deep Learning Research Engineer

GAIPS Research•

Artificial Intelligence (AI)

Feb 2019 - Oct 2024 · 5y 8m

Designed, implemented, trained, tested, and deployed state-of-the-art deep learning architectures, including Actor-Critics, DQNs, and LLMs, using convolutional, recurrent, and attention-based mechanisms for feature extraction across a wide range of tasks.
Deployed models to cloud production environments on platforms such as Google Cloud, Amazon AWS, and Slurm HPC distributed computing clusters.
Monitored the continual performance of deployed models using tools like MLFlow, Sacred, and Weights & Biases.
Assembled the company’s HPC Slurm cluster.
Led five research teams as first author, publishing a research paper for each in top-tier AI venues, including AAAI, IJCAI, ECAI, the Artificial Intelligence Journal, and PLoS One Journal.
Presented AI research at top-tier international conferences such as AAAI, IJCAI, and ECAI.
Secured two competitive funding grants, one from the U.S. Air Force Office of Scientific Research and another from the Portuguese Foundation for Science and Technology (FCT).
Received the Best Paper award for the project “Helping People On The Fly: Ad Hoc Teamwork for Human-Robot Teams.”

Docker

Python

Data Science

Joomla

NumPy

30+

Software Engineer

Thales•

Aerospace and Defense

May 2018 - Jan 2019 · 8m

Reduced technical debt and increased overall test coverage of the Top Sky Tower solution, a tool for air traffic controllers to manage electronic strips.
Implemented and tested critical security detection systems.

Java

C++

WPF

Software Engineer

IST IT Department•

Information Technology (IT) and Services

Mar 2017 - Mar 2018 · 1y

Trained a Convolutional Neural Network to classify valid identity card images.
Implemented software for automatic and periodic backups of the university’s records to the AWS cloud.
Re-implemented legacy software using modern technologies such as Scala and Kotlin.

Java

Python

AWS S3

Scala

Kotlin

Software Engineer

DV Trading LLC•

Stock Trading

Jul 2016 - Sep 2016 · 2m

• Implemented graphical user interfaces using WPF and .NET for the trading team • Refactored and optimized code in several legacy projects, increasing overall performance of proprietary trading tools by up to 30%

Java

C++

.NET

WPF

Laravel Developer

Systems Group•

Information Technology (IT) and Services

Feb 2016 - Jun 2016 · 4m

Designed and developed a website for the Trainees project - matchmaking companies and near-graduates from the Portuguese ESHTE

PHP

Laravel

MySQL

MariaDB

MongoDB

Docker

Engineering excellence

All Deep Learning Research Engineers who have applied to Proxify are scored from 0 to 300 on engineering excellence, one of the five parameters we evaluate. This score reflects engineering excellence only, based on interviews, take-home assignments, live coding sessions, and/or on-the-job performance reviews. The curve shows how all evaluated Deep Learning Research Engineers are distributed across that range, where our acceptance threshold for this parameter sits, and where João stands.

João
Score 180 · Top 14% of engineers

See vetting process

Portfolio

Highlighted by João

Sep 2017

Multi-Task Learning & Catastrophic Forgetting in Continual Reinforcement Learning

This project investigates two hypothesis regarding the use of deep reinforcement learning in multiple tasks. The first hypothesis is driven by the question of whether a deep reinforcement learning algorithm, trained on two similar tasks, is able to outperform two single-task, individually trained algorithms, by more efficiently learning a new, similar task, that none of the three algorithms has encountered before. The second hypothesis is driven by the question of whether the same multi-task deep RL algorithm, trained on two similar tasks and augmented with elastic weight consolidation (EWC), is able to retain similar performance on the new task, as a similar algorithm without EWC, whilst being able to overcome catastrophic forgetting in the two previous tasks. We show that a multi-task Asynchronous Advantage Actor-Critic (GA3C) algorithm, trained on Space Invaders and Demon Attack, is in fact able to outperform two single-tasks GA3C versions, trained individually for each single-task, when evaluated on a new, third task—namely, Phoenix.

We also show that, when training two trained multi-task GA3C algorithms on the third task, if one is augmented with EWC, it is not only able to achieve similar performance on the new task, but also capable of overcoming a substantial amount of catastrophic forgetting on the two previous tasks.

Duration1y 1m