Jonathan Mamou
Deep Learning And NLP Researcher at Intel Corporation
Israel
Summary
🤩
Rockstar🎓
Top SchoolJonathan Mamou is a Deep Learning and NLP researcher with a PhD in Computer Science and over a decade of hands-on experience driving language-model innovations at Intel. He has contributed production-grade improvements to flagship open-source projects such as Hugging Face Transformers—implementing speculative decoding strategies and assistant confidence mechanisms—and developed the NP2vec model inside IntelLabs' nlp-architect for term set expansion and model optimization. Based in Israel, Jonathan blends rigorous academic training with practical ML engineering, producing tests, training scripts, and iterative enhancements that move research prototypes toward usable tooling. His work shows a pattern of improving both core algorithms and developer ergonomics, from token scheduling heuristics to validation and logging fixes. Colleagues can expect a researcher who combines deep theoretical knowledge with pragmatic implementation skills and a track record of shipping reliable, community-facing contributions.
10 years of coding experience
Doctor of Philosophy (PhD), Computer Science, Doctor of Philosophy (PhD), Computer Science at Université Paris Sud (Paris XI)
Github Skills (21)
pytorch10
language-model10
word2vec10
python10
machine-learning10
nlpjs10
deep-learning10
natural-language-processing10
transformer10
nlp10
hub9
fasttext9
jax8
flax8
pre-trained-model8
Programming languages (2)
Github contributions (5)
IntelLabs/nlp-architect
Mar 2018 - Feb 2019
A model library for exploring state-of-the-art deep learning topologies and techniques for optimizing Natural Language Processing neural networks
Role in this project:
ML Engineer
ML EngineerContributions:50 commits, 1 push, 2 branches in 11 months
Contributions summary:Jonathan primarily focused on the development and implementation of an NP2vec model within the nlp-architect repository. Their initial commit introduced the core components of the NP2vec model, including training and saving functionalities. Subsequent commits addressed issues such as log forging, code documentation, and added enhancements like validation, showing a continuous effort to improve the model's functionality and usability. Their contributions involved creating training scripts and demonstrating the model's application in term set expansion.
nlunatural-language-understandingbertlanguage-processingstate-of-the-art
huggingface/transformers
Nov 2019 - Apr 2025
🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.
Role in this project:ML Engineer
ML EngineerContributions:19 reviews, 11 PRs, 60 comments in 5 years 5 months
Contributions summary:Jonathan primarily contributed to the development and refinement of the speculative decoding and assisted generation functionalities within the transformers library. This included implementing the "heuristic" and "adaptive" token scheduling strategies, and incorporating a confidence threshold mechanism for the assistant model. These changes involved modifications to configuration files, candidate generator logic, and stopping criteria, ultimately aiming to improve the efficiency and performance of the speculative decoding process. The user also added tests to ensure the correct behavior of these features.
pythonbertspeech-recognitionstate-of-the-artflax
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.