Skip to content

ML Engineer · MLOps · LLMs · Computer Vision

I design, train, and ship machine-learning systems that hold up in production.

I’m Youssef Elghoudani — a machine learning engineer working across LLM applications, computer vision, and the MLOps that turns a notebook into a reliable, monitored service.

View selected workGet in touch
✻ Youssef

› whoami

youssef elghoudani · ml engineer

› cat focus.txt

llm apps · computer vision · mlops

› ./ship --target production

✓ trained, served, monitored — and holding up

›

Certifications issued by IBM, Microsoft, Google, Columbia University, Amazon Web Services, Corporate Finance Institute, University of Illinois Urbana-Champaign, University of California, Davis, University of London.

IBM26Microsoft15Google10Columbia University6Amazon Web Services5Corporate Finance Institute5University of Illinois Urbana-Champaign5University of California, Davis4University of London1

Portfolio assistant

Scrolling optional.

Everything further down this page — the projects, the stack, the experience — is one question away.

Or scroll. That works too
Open full screen

About

Two disciplines, and a preference for problems that matter.

I’m Youssef Elghoudani, a Machine Learning Engineer with a dual background in computer science and statistics, working mainly in deep learning. The statistics half is what stops me shipping a number I can’t defend; the engineering half is what gets it in front of someone.

I’ve applied that to medical imaging, where a classifier supports a specialist’s judgement rather than replacing it, and to organisations whose own data was too fragmented to answer their questions. What both taught me is that the hard part is rarely the model.

structure entropy
Focus
LLMs · CV · MLOps
Experience
Production ML systems
Approach
Measured & maintainable

Skills & stack

The toolkit I reach for, grouped by where it lives in the stack.

LLMs & Generative AI

  • Agents & tool use
  • LoRA · QLoRA fine-tuning
  • DPO · RLHF
  • Quantization (GGUF · AWQ)
  • LLM-as-judge evals
  • vLLM serving

Deep Learning

  • PyTorch
  • Hugging Face Transformers
  • Datasets · Accelerate
  • Transformer internals
  • Attention & KV cache
  • DDP · FSDP · DeepSpeed
  • Mixed precision
  • CNNs · RNNs · autoencoders

Computer Vision

  • YOLO · Detectron2
  • Vision Transformers (ViT · Swin)
  • CLIP & multimodal
  • Self-supervised (DINOv2)
  • Diffusion & generative vision
  • Object tracking · OCR
  • ONNX · TensorRT · quantization
  • Real-time video inference

Engineering & MLOps

  • Python · SQL
  • FastAPI
  • Docker · Kubernetes
  • MLflow · Weights & Biases
  • CI/CD for ML
  • Monitoring & drift

Selected work

A few systems I’ve built, and the results they delivered.

01

Automated Leukocyte Classification

I worked with the R&D arm of one of the largest medical-biology groups in Alsace — more than 40 labs — building an in-house model that classifies white blood cells from blood-smear images. Trained on 39,037 clinician-validated cells across 12 classes, it outperformed a re-implemented top-3 competition solution on the same test split, and beat vision backbones ten times its size while running several times faster.

  • PyTorch
  • YOLO-cls
  • SigLIP2 · DINOv2
  • Medical imaging
Read the case study
95.8%
macro F1 across 12 cell classes
02

The Link

I worked with one of the largest residential landlords in Alsace, where every department held its own data, used it for its own tasks, and could not see past its own boundary — with basic reporting rented from outside vendors. Before any modelling was possible, I spent weeks mapping what each department held and what it needed, and joining those sources into one base covering the whole estate.

  • Python
  • pandas
  • SQL
  • Data integration
Read the case study
13
variables catalogued into one base
03

The Value

Chapter two of the same engagement: with the data finally joined, modelling what actually sets rent per square metre.

  • scikit-learn
  • SciPy
  • Web scraping
  • Forecasting
Project complete — write-up in progress
60%
of predictive weight in one financing rule
04

AI Assistant for Cross-Continental Freight

An AI chatbot for a large startup moving cargo between Europe and the Gulf states, built around the way freight actually gets booked, tracked and chased across two regions and several time zones.

  • LLMs
  • Conversational AI
  • Logistics
In progress
Report coming soon — the work is still running

Competitions

Ranked in public.

I’m taking competitions more seriously from here on, including for the lab I work with. Each one gets added here as it happens — the platform, the problem, and where it finished.

6th place
Kaggle

CSIRO — Image2Biomass Prediction

Predicting biomass from imagery, run by Australia's national science agency.

Computer VisionRegression
Top 10
Kaggle

AI Mathematical Olympiad — Progress Prize 2

Solving olympiad-level mathematics problems with open models under a fixed inference budget.

LLMsMathematical Reasoning

Certifications

If you care about certifications.

All 77 of them, here — 9 complete programmes and the courses inside them, across 9 issuers. Open any one for its credential ID and a link to its Coursera verification page.

Research

What I’m reading.

The papers I’m working through at the moment. It changes with whatever I’m building — each project pulls a different set — so I edit this as the work moves.

arXiv · 2023

DINOv2: Learning Robust Visual Features without Supervision

The self-supervised backbone I benchmarked against on the leukocyte work. Worth understanding why features learned without labels transfer to a domain — blood-smear microscopy — that looks nothing like the pre-training data.

Self-supervisedVision BackbonesRead paper →
arXiv · 2025

SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features

The other backbone in that comparison, and the one that came closest. Reading it properly to understand what the sigmoid loss buys over a contrastive objective at this scale.

Vision-LanguageRepresentation LearningRead paper →
arXiv · 2026

Are General-Purpose Vision Models All We Need for 2D Medical Image Segmentation? A Cross-Dataset Empirical Study

The same question I ran into on the leukocyte work, asked properly and across datasets: do general-purpose backbones actually beat models built for the medical task? My own benchmark said no — the 431M-parameter foundation model finished last. Reading this to find out how far that generalises.

Medical ImagingSegmentationRead paper →
arXiv · 2026

Uncertainty-Aware Vision-Language Segmentation for Medical Imaging

Uncertainty is the missing half of a clinical classifier. A model that knows when it is unsure is worth more to a biologist than one that is slightly more accurate and silent about it — which is exactly where the cell-classification work needs to go next.

Medical ImagingUncertaintyRead paper →
arXiv · 2026

Bootstrapping Vision-Language Model for Hysteroscopic Surgical Scene Segmentation

Vision-language models pushed into a real clinical setting, where the imagery looks nothing like the web data they were pre-trained on. The same domain gap I hit taking general backbones to blood-smear microscopy.

Vision-LanguageClinical ImagingRead paper →
arXiv · 2022

ReAct: Synergizing Reasoning and Acting in Language Models

Interleaving reasoning with tool calls. Directly relevant to the freight assistant and to how the tool layer behind Prometheus on this site is structured.

LLM AgentsTool UseRead paper →
arXiv · 2023

Efficient Memory Management for Large Language Model Serving with PagedAttention

The vLLM paper. Serving is where most LLM projects quietly become too expensive to run, and this is the clearest account of why the KV cache is the thing that decides that.

InferenceServingRead paper →

Contact

Let’s build something that makes it to production.

Open to ML engineering roles (CDI). The fastest way to reach me is email — I usually reply within a day.

Reach out

Leave a brief message

I’m looking for…

© 2026 — Designed & built by Youssef Elghoudani