Teodor Poncu
Member Of Engineering at University POLITEHNICA of Bucharest
Romania
Summary
🤩
Rockstar🎓
Top SchoolTeodor Poncu is a machine learning researcher and engineer with a decade of experience specializing in unsupervised, self-supervised, and generative modeling, currently focused on large-scale LLM pre-training and reinforcement learning at poolside. He holds an active research role at University POLITEHNICA of Bucharest, leads a small medical-imaging research group, and is pursuing a PhD in unsupervised representation learning. Teodor has shipped production stereo-matching and depth pipelines at Meta and contributed notable dataset and stereo utilities to the widely used pytorch/vision repository. His applied work spans histopathology diagnostics, autonomous-driving domain adaptation, and practical CodeGen/agent research that achieved strong benchmarks versus GPT-4 in low-resource settings. Comfortable bridging research and engineering, he builds end-to-end training and inference systems while mentoring students and teaching advanced AI topics. A subtle throughline in his career is reducing distribution-shift errors via augmentation-free and teacher-student strategies, translating academic ideas into robust, deployable models.
10 years of coding experience
6 years of employment as a software developer
POLITEHNICA București National University for Science and TechnologyBachelor of Science - BS, Computer Science, Bachelor of Science - BS, Computer Science at Universitatea „Politehnica” din București
Github Skills (5)
pytorch10
machine-learning10
computer-vision10
datasets10
python9
Programming languages (2)
Github contributions (5)
pytorch/vision
Jul 2022 - Sep 2022
Datasets, Transforms and Models specific to Computer Vision
Role in this project:
ML Engineer
ML EngineerContributions:92 reviews, 147 commits, 28 PRs in 2 months
Contributions summary:Teodor primarily focused on enhancing the `pytorch/vision` repository, which specializes in computer vision. Their main contributions involved integrating and improving datasets for stereo matching tasks, including the addition of new datasets like `SintelStereo`, `InStereo2k`, `FallingThingsStereo`, `ETH3DStereo`, and `CREStereo`. They refactored existing code related to stereo matching and adapted testing utilities to accommodate the new datasets and configurations.
pytorchvisiondeep-learningdatasetcomputer-vision
TeodorPoncu/vision
Jul 2022 - Sep 2022
Datasets, Transforms and Models specific to Computer Vision
Contributions:77 pushes, 5 branches in 1 month
imagenetclassificationdata-profilingdata-preprocessingvision
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.