Adil Eddarif — AI Engineer & AI Consultant in Morocco
AI engineer building production medical AI — retrieval-augmented clinical assistants and computer-vision triage pipelines that ship behind real safety constraints.
AI engineer working on production medical AI systems at the Cheikh Zaïd Foundation in Rabat, Morocco. I build retrieval-augmented generation (RAG) assistants and computer-vision pipelines, and I care most about the part most demos skip: evaluation, failure modes, and the humans who have to trust the output. Alongside that I ship satellite computer-vision pipelines for GeoAP, and previously did AI and data science at Attijariwafa Bank and computer vision at Arteka.
I work as an AI engineer and AI consultant: building the systems, and advising the teams adopting them.
What I actually ship
Raised emergency-case recall from 50% to 100% on a clinical RAG assistant by finding a safety-routing bug with an LLM-as-judge evaluation harness.
Shipped a two-stage chest X-ray triage pipeline (ResNet50 + YOLOv8) at 96.93% accuracy and 99.71% AUC on a 522-image held-out test set.
Built a 1,000-question golden evaluation dataset with independent faithfulness and hallucination scoring.
Replaced heuristic financial forecasting rules at Attijariwafa Bank with supervised models at 0.90 R², tracked end-to-end in MLflow.
Delivered a satellite computer-vision pipeline for GeoAP across 10 object classes — YOLO11m detection at 0.836 mAP50 and a fine-tuned SAM1 ViT-H segmenter at 0.804 mIoU — exported as GIS-ready GeoJSON.
Cut SAM1 ViT-H fine-tuning from 88 minutes to 5 per epoch on a single T4 GPU by pre-computing and caching image embeddings across 9,958 tiles.
Reached >99% recognition accuracy on Tifinagh script OCR across a 25,740-image dataset.
Explore this site
About — Background, education, languages, and how I work.
Experience — Roles across medical AI, satellite CV and banking, with measured outcomes.
Skills — AI, MLOps, and software engineering stack.
Projects — Case studies with architecture, trade-offs, and results.
Blog — Medical AI, RAG systems, enterprise AI agents, and production ML.
Contact — How to reach me and what I am available for.
Latest writing
Two-Stage Chest X-Ray Triage: ResNet50 + YOLOv8 at 96.9% Accuracy — A classifier tells you something is there. A detector tells you where. Neither alone is enough for clinical triage. Here is the two-stage chest X-ray pipeline I built, why the architecture is deliberately unfashionable, what the preprocessing actually contributes, and how to read the 96.93% accuracy and 99.71% AUC honestly.
Enterprise AI Agents: The Architecture That Survives Contact With Compliance — Most AI agent architectures are designed for capability and then retrofitted for control, which is why so few reach production inside a real company. Here is what changes when an agent has to pass an audit — permissions, data boundaries, cost governance, human approval, and the evidence trail — and how to build for it from the start.
Why Medical AI Fails in Production — and Why Data Quality Decides It — A medical AI model can score 96% on a held-out test set and still be unsafe in a hospital. The gap is almost never the architecture — it is the data. Here are the five data-quality failures that decide whether clinical AI survives contact with production, and how to catch each one before a patient does.
RAG vs Fine-Tuning vs Prompt Engineering: How to Actually Choose — The three ways to make a language model do what you want are constantly compared as if they were competitors. They solve different problems. Here is the decision rule, what each one actually costs, why most teams reach for fine-tuning when they needed retrieval, and how to tell which situation you are in.
Contact
Available for AI engineering and consulting work. Based in Rabat, Morocco.