Valeriu Lacatusu
Senior Machine Learning Engineer at Meta
France
Summary
👤
Senior🎓
Top SchoolValeriu Lacatusu is a Senior Machine Learning Engineer based in France with 14 years of experience building production-grade ML systems in computer vision, audio/MIR, and visual document understanding. He combines research-oriented R&D with hands-on engineering, having led ML teams at Xtramile and currently driving projects at Meta. Valeriu has a strong background in distributed systems and scalable deployment—designing grid computing platforms for media recognition and end-to-end APIs for video indexing. An active open-source contributor, he improved performance and torch-compile compatibility in Facebook Research’s xformers and enhanced database-specific optimizations in linq2db. His work spans low-level optimization (CUDA kernels, multiprocessing) to high-level dataset and training integrations (RetinaNet generators), reflecting a rare mix of systems, ML, and data engineering skills. Colleagues describe him as an independent researcher who turns experimental ideas into robust, production-ready solutions.
14 years of coding experience
12 years of employment as a software developer
International High School of BucarestLycée Louis-le-Grand
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (26)
pytorch10
oracles10
python10
linq10
net10
databases10
machine-learning10
dotnet10
kernel10
asp-net10
keras10
oracle10
tensorflow10
sql10
dotnet-core10
Programming languages (4)
Github contributions (5)
linq2db/linq2db
Dec 2013 - Oct 2017
Linq to database provider.
Role in this project:
Back-end Developer & Database Engineer
Back-end Developer & Database EngineerContributions:27 commits, 5 PRs, 42 comments in 3 years 10 months
Contributions summary:Valeriu primarily focused on enhancing the Oracle database provider within the linq2db project. Their contributions include implementing bulk copy functionality, refining date/time handling, and incorporating support for the `NVARCHAR2` data type. Furthermore, the user refactored code related to sequence value retrieval to improve efficiency and data mapping accuracy, demonstrating proficiency in both data access and database-specific optimizations. They also made changes to fix a SQLite table name issue.
sqlserverasp-net-core-web-apisqlcesap-hanaentity-framework
fizyr/keras-retinanet
Jan 2018 - Nov 2018
Keras implementation of RetinaNet object detection.
Role in this project:ML Engineer
ML EngineerContributions:30 commits, 21 PRs, 125 comments in 9 months
Contributions summary:Valeriu primarily contributed to the development and enhancement of the object detection capabilities within the Keras RetinaNet project. Their work includes adding a new generator for the Google Open Images dataset, incorporating support for the Kitti dataset, and implementing the MobileNet backbone. Furthermore, the user made modifications to training scripts, added new arguments, and adjusted the handling of dataset versions and filtering to provide better usability and flexibility to the object detection pipeline.
deep-learningretinanetcomputer-visionobject-detectiontensorflow
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.