Research
From Darija NLP to fetal ultrasound — peer-reviewed, cited, published.
SalamDarija
AI Solution Architect
I take messy real-world data — ultrasound video, bank records, spoken Darija — build models around it, evaluate them honestly, and ship them to production.

Anass Allak
Speaking on Darija speech recognition
Now
Full Stack Data Scientist at Deepecho
Fetal ultrasound AI, from raw video to production.
Based in
Casablanca, Morocco
00:00:00
Overfitting playground
Click to add data, click a dot to remove it. Then crank the degree.
A straight line. Simple, honest, maybe too simple.
0+
Years in industry
0+
Projects shipped
0
Research papers
0+
Technologies
Explore
Jump to any project or section from the keyboard. And psst — try typing salam.
Real data is messy. I build models that survive it — for doctors, banks & insurers and people who speak Darija.
I'm a Full Stack Data Scientist working across machine learning, data engineering and applied AI, and a PhD candidate in Machine Learning at the National Institute of Statistics and Applied Economics (INSEA). My work spans clinical fetal ultrasound AI, speech recognition and language models for Moroccan Darija, fraud, credit-risk and churn models for banks and insurers, and the data pipelines and serving infrastructure that keep it all running.
How I work
From Darija NLP to fetal ultrasound — peer-reviewed, cited, published.
Models in production, not just notebooks. FastAPI + Triton + ONNX on GCP.
Video to dataset to model to serving. Owned end-to-end.
Site-generalization eval, failure analysis, honest metrics.
Any modality
I architect ML systems across every data modality: structured tabular, text and language, voice and audio, image, video. The modeling, data engineering, and MLOps patterns are the same — the signal just lives in a different shape.
fraud, credit risk, churn
Moroccan Darija ASR, Arabic NLP, LLM pipelines
ASR, speaker ID, acoustic diagnostics
fetal ultrasound, DICOM, medical CV
ultrasound sweeps, temporal CV, action recognition
By the numbers
0+
Citations
Google Scholar, across venues
0
Peer-reviewed papers
Including NeurIPS workshops
0
h-index
Scholar, lifetime
Selected work
Hover to trace links · click a project · drag nodes around
Experience
Deepecho
Building fetal ultrasound AI end to end — from raw video and annotation to clinical models, evaluation and production serving.
INSEA — National Institute of Statistics and Applied Economics
Doctoral research on NLP and speech recognition for Arabic dialects.
Aiox-labs
Led AI and data projects for major Moroccan banks, insurers and infrastructure operators.
Aiox-labs
Voice recognition and generation for the Moroccan Arabic dialect.
ENSA Tétouan — Telecommunications Laboratory
Final-year project: an intelligent OFDM radio transmission simulation platform.
Research
Full list on Google Scholar · ORCID
Toolbox
Filter by domain to see what powers each kind of project.
Teaching · Professor
As a professor, I design and teach these courses — taking students from fundamentals to the tools I use in production.
Practical analysis workflows: exploratory analysis, statistical inference, visualization and end-to-end analytical pipelines.
From NLP fundamentals to transformers, LLM fine-tuning and low-resource language processing — with a focus on Arabic dialects.
Distributed processing, cloud data pipelines and scalable infrastructure for large datasets in production.
Live from GitHub & Hugging Face
Contact
Have a data problem, a research idea or a role in mind? Send a message and I'll get back to you.
[email protected]Casablanca, Morocco
Speech recognition toolkit and open dataset for Darija and other underrepresented dialects.
Darija speech-to-text model trained on DVoice, published in the SpeechBrain model hub.
The first open-source BERT language model for Moroccan Darija.
DarijaBERT for Darija written in Latin script (Arabizi), as people actually type it online.