Experience Projects Publications Contact Photography

Senior ML Research / Software Engineer · Paris, France

YASSIR El Mesbahi

Developing cutting-edge machine learning models and systems to solve real-world problems — ML Research Engineer · Software Engineer · Deep Learning · Generative AI · LLMs · Biotechnology
Yassir El Mesbahi
Core Competencies
LLMs & Foundation Models97%
Distributed Training (FSDP/DeepSpeed)95%
Multimodal & BioML90%
High-performance Pipelines93%
Systems Engineering (C/C++)85%
Languages
EnglishFluent
FrançaisNative
العربيةNative
Education
2019 – 2021
M.Sc. Artificial Intelligence
Université de Montréal — MILA
2009 – 2013
M.Sc. Computer Science & Applied Mathematics
Grenoble INP — ENSIMAG, France
LLMsGenerative ModelsBioMLDistributed TrainingFSDP · DeepSpeedMulti-GPUPyTorchDrug DiscoveryNLPFoundation Models LLMsGenerative ModelsBioMLDistributed TrainingFSDP · DeepSpeedMulti-GPUPyTorchDrug DiscoveryNLPFoundation Models
Experience
Where I've worked
2023 – 2025
Valence Labs
(Recursion Pharmaceuticals)
Montréal, Canada
Full Time
ML Research Engineer — Bio-modality Team
  • Built large-scale training infrastructure for transcriptomics and multimodal biological models (PyTorch + DeepSpeed + Slurm), scaling across multi-node clusters of 32–64 GPUs.
  • Built research codebases used by biology and physics teams, improving reproducibility and accelerating experimentation cycles.
  • Co-authored four papers (NeurIPS, ICML, Nature Biotechnology) on multimodal learning, perturbation prediction, and generative models for drug discovery.
  • Implemented GPU-efficient training pipelines (FSDP, mixed precision, streaming data loaders), reducing memory footprint and compute cost significantly.
2023
New York University
Tandon School of Engineering
Montréal, Canada
Part Time
ML Research Associate — Finance & Risk Engineering
  • Research assistant with Prof. Amine Aboussalah on data-driven modeling and financial ML.
  • Mentored graduate students; teaching assistant for Data-Driven Dynamical Systems.
2021 – 2023
Huawei Technologies
NLP Team
Montréal, Canada
Full Time
ML Research Engineer
  • Efficient pre-training methods and LLM compression (pruning, quantization, distillation) on NLP/NLU benchmarks.
  • Converted prototypes into production services: Arabic punctuation restoration deployed in Huawei products.
  • Co-authored two publications (NeurIPS, CoNLL).
2017 – 2019
Teradata
Unity Load Team
Toronto, Canada
Full Time
Software Development Engineer II
  • Multi-threaded C components for petabyte-scale enterprise data systems at high-speed ingestion rates.
  • Deadlock-prevention features and modular unit testing infrastructure across the product line.
2013 – 2017
Dassault Systèmes
Material Modeler Team
Paris, France
Full Time
Software Development Engineer
  • Software design, testing and maintenance in C++.
  • Middleware enabling xPDM data exchange with partners — strategic inter-company integration project.
Open Source
GitHub Projects
github.com/yaemsi ↗
01
Python
txpert
Research codebase for TxPert — OOD transcriptomic perturbation prediction leveraging biochemical relationships. Nature Biotechnology submission.
PyTorchDeepSpeedscRNA-seq
Nature BiotechnologyView
02
Python
fm-benchmarking
Benchmarking suite for transcriptomics foundation models across perturbation analysis tasks. Reproducible pipeline — NeurIPS 2024 AIDrugX.
HuggingFacePyTorchSlurm
NeurIPS 2024View
03
Python
cross-modal-kd
Knowledge distillation and data augmentation bridging morphological imaging and transcriptomics representations. ICML 2025.
FSDPMixed PrecisionPyTorch
ICML 2025View
04
Python
llm-compression
Compressing decoder-based LMs via pruning, quantization, and distillation. Research at Huawei — NeurIPS 2021 ENLSP.
TransformersQuantizationDistillation
NeurIPS 2021View
05
Python · Shell
ml-training-infra
Reusable templates for large-scale distributed ML: FSDP configs, Slurm scripts, streaming data loaders, and mixed-precision recipes.
SlurmFSDPDeepSpeed
UtilitiesView
Research
Papers &
Publications
01
TxPert: Leveraging Biochemical Relationships for Out-of-Distribution Transcriptomic Perturbation Prediction
Nature Biotechnology
Submitted · 2025
2025
Read ↗
02
A Cross Modal Knowledge Distillation & Data Augmentation Recipe for Improving Transcriptomics Representations through Morphological Features
ICML 2025
2025
Read ↗
03
Benchmarking Transcriptomics Foundation Models for Perturbation Analysis: one PCA still rules them all
NeurIPS 2024 · AIDrugX Workshop
2024
Read ↗
04
SAFE Setup for Generative Molecular Design
NeurIPS 2024 · AI4Mat Workshop
2024
Read ↗
05
On the Utility of Enhancing BERT Syntactic Bias with Token Reordering Pretraining
CoNLL 2023
2023
Read ↗
06
A Short Study on Compressing Decoder-Based Language Models
NeurIPS 2021 · ENLSP Workshop
2021
Read ↗
Beyond the lab
Photography
Portfolio
AI Assistant
Ask me anything

An AI briefed on my full background. Ask about experience, papers, projects, or availability.

Contact
Let's build
something.

Open to research collaborations, consulting, speaking engagements, and anything genuinely interesting.

Emailyassir.elmesbahi@outlook.com LinkedInyassir-el-mesbahi Google ScholarView publications Phone+1 514-346-9319