AI/ML Engineer · Algeria

I build the model.
Then I build
the product.

Applied deep learning for clinical and diagnostic use cases — imaging, sequential decision-making, tabular health data — and the full-stack engineering that ships it.

Explore work ↗ Download CV

Model architecture

99%

Brain tumor accuracy

BBC

4Tech featured

Apps in production

Selected work

From model
to product.

Projects span clinical AI, computer vision, reinforcement learning, and full-stack products. Accuracy figures are evaluation results, not claims of clinical deployment.

Featured · Medical computer vision · BBC 4Tech

HemaVision

Automated hematology platform replacing manual microscopy for routine blood analysis. YOLOv8 detection, U-Net segmentation, and a disease layer flagging cell morphology tied to leukemia, malaria, and anemia. Validated by a clinical hematologist before being featured on BBC News Arabic's 4Tech program.

YOLOv8 U-Net OpenCV PyTorch 97% accuracy 45 min → 3 min

Watch BBC coverage ↗

From smear image to clinical output

01 · INPUT

Detection YOLOv8 locates cells in dense smear images

02 · PROCESS

Segmentation U-Net isolates overlapping cell regions

03 · OUTPUT

Classification RBC / WBC subtypes + disease flags

Computer vision pipeline Clinician-validated

01 Medical computer vision

AIRM — Brain MRI Classification

EfficientNet-B7 reaching 99% across four tumor categories with a full DICOM pipeline and a radiologist-validated PyQt5 interface built for real clinical workflow integration.

Result

99% four-class accuracy. DICOM pipeline tested with hospital MRI scanners.

02 Reinforcement learning · Healthcare

Medical Treatment DRL

ICU treatment timing agent on MIMIC-III. A2C reached 99.5% clinical appropriateness with +76–145 reward improvement. Rule-based safety filter lifted PPO/DQN appropriateness by 40 points.

Result

99.5% appropriateness. 26-dimensional Gym environment modeling temporal lab trends.

03 Full-stack product

SpecMob

Phone comparison and discovery platform. Next.js + FastAPI + async Postgres, trigram typo-tolerant search, tiered caching. Product copy generated by Gemini with a deterministic fallback — AI layer never blocks rendering.

Live product

Full-stack from nothing to production: frontend, API, database, AI copy layer.

04 Applied ML · Research thesis

My Daily Health

Multi-disease diagnostic platform across five domains. Systematic evaluation of 12 architectures with stratified cross-validation. Secure clinical workflow, sub-200ms inference. ICSTEM 2023 Outstanding Presentation Award.

Result

90–99% accuracy across five disease domains on M.Sc. thesis evaluation.

05 Healthcare Cost Prediction Conv1D · SHAP · R² 0.88 View ↗

06 Deep RL for Crypto Trading PPO · A2C · Negative result View ↗

07 Day Tracker Next.js · PostgreSQL · Web Push View ↗

08 Git-Backed CMS Next.js · GitHub Contents API Visit ↗

About

Research work makes me careful about what a model actually knows. Systems work makes me care about latency, failure modes, and what happens when the happy path breaks.

I work across applied deep learning for clinical and diagnostic use cases — imaging, sequential decision-making, tabular health data — and the full-stack engineering that ships a model as a product. The two feed each other.

My work spans brain tumor classification, automated blood cell detection, and ICU treatment-timing agents. Each project is pushed toward something a clinician or user could plausibly use, not just a benchmark score.

On the systems side: Next.js and React on the frontend, FastAPI or Flask on the backend, Postgres for storage. I've taken a full product from nothing to production and keep several Postgres-backed apps running.

Core focusMedical AI · Applied systems

Technical rangeCV · DL · RL · Web

EducationM.Sc. Embedded Systems

Based inAlgeria

Contact

Open to research collaborations and ML engineering roles.

Reach out if you're working in medical AI, need an applied ML engineer, or need someone who can take a model all the way to a shipped product.

Email me ↗

I build the model.Then I buildthe product.

From modelto product.

HemaVision

Open to research collaborations and ML engineering roles.

I build the model.
Then I build
the product.

From model
to product.