Case studies

Every project, start to finish.

Research prototypes and production-minded systems across generative AI, computer vision, NLP and MLOps. Filter by domain.

Knowledge-Guided Diffusion for Medical Imaging cover

Generative AI
Computer Vision

2026

Knowledge-Guided Diffusion for Medical Imaging

Conditioning diffusion models with textual and spatial medical knowledge

Research project investigating how medical knowledge can be incorporated into diffusion-based image generation. The approach combines Stable Diffusion with LoRA fine-tuning and knowledge-guided conditioning using textual clinical information and spatial pathology annotations. Conducted as part of a Master's thesis at CIAD Lab, UTBM, France.

Python
PyTorch
Stable Diffusion
LoRA
ControlNet
+2

LallaCare — Multimodal Breast Cancer Risk Assessment cover

Computer Vision
Data Science
Generative AI

2025

2nd Prize — MoroccoAI Summer School Hackathon

LallaCare — Multimodal Breast Cancer Risk Assessment

Hackathon prototype for multimodal breast cancer risk assessment

A hackathon prototype exploring multimodal breast cancer risk assessment by combining thermal imaging with structured clinical information. The system uses computer vision features from thermal images alongside clinical variables to produce a machine-learning-based risk score. 🏆 2nd Prize — MoroccoAI Summer School Hackathon 2025.

Python
PyTorch
Scikit-learn
OpenCV
LLMs
+2

NLP
Generative AI

2025

Historical Figures Chatbot

Conversation with history's greatest thinkers, in character

A conversational AI application that lets users chat with historical figures — Socrates, Cleopatra, Tesla, Rumi and others — powered by LLaMA 3.1 via the NVIDIA NIM API. Each figure is grounded in a detailed character profile built from their documented works, speech patterns and historical context.

Python
LLaMA 3.1
NVIDIA NIM API
LangChain
Streamlit
+1

MemoryDiffusion — Training-Free Personalisation cover

Generative AI
Computer Vision

2026

MemoryDiffusion — Training-Free Personalisation

Experimental investigation of training-free personalisation for diffusion models

An experimental implementation exploring training-free personalisation for diffusion models using an external visual memory. Reference images are encoded and retrieved at generation time rather than baked into the model weights.

Python
PyTorch
Diffusion Models
Embeddings
CLIP
+2

DataID — Agentic Data Science Workflow cover

Generative AI
Data Science
MLOps

2025

DataID — Agentic Data Science Workflow

A multi-agent prototype for automating parts of the data science workflow

A multi-agent system that orchestrates several stages of a data science workflow, from exploratory analysis and preprocessing to model training and explainability. Built as a prototype to demonstrate agentic automation of routine data science tasks.

Python
LLMs
AI Agents
AutoML
SHAP
+3

AtlasAI Morocco — AI Earth Observation Platform cover

Computer Vision
Data Science

2025

AtlasAI Morocco — AI Earth Observation Platform

AI-powered geospatial exploration of environmental indicators in Morocco

An AI-powered geospatial platform exploring the use of satellite imagery and Earth observation data for environmental monitoring in Morocco. Implements vegetation monitoring, water-resource tracking and land-cover analysis over time.

Python
AI Agents
Remote Sensing
Geospatial AI
PyTorch
+2

CertifyAI — Blockchain Certification for AI Models cover

MLOps
Data Science

2025

CertifyAI — Blockchain Certification for AI Models

Claiming ownership of AI model weights using blockchain and cryptographic hashing

A system for establishing and verifying ownership of machine learning model weights using SHA-256 hashing, IPFS content addressing and an Ethereum smart contract registry. Each model artefact is fingerprinted and the hash is recorded on-chain, so ownership can be claimed and verified independently.

Ethereum
Solidity
Smart Contracts
IPFS
SHA-256
+3

LLM Security — Vulnerability Evaluation cover

NLP
Generative AI

2025

LLM Security — Vulnerability Evaluation

Measuring how language models fail under adversarial pressure

A study of large language model risks through the analysis of prompt injection, jailbreak attacks and unsafe content generation.

Python
LLM Security
Prompt Engineering
AI Safety
Hugging Face
+1

Scale-Sell — Intelligent Fruit Recognition for Retail cover

Computer Vision
Generative AI

2026

Scale-Sell — Intelligent Fruit Recognition for Retail

Vision-based produce recognition and weight estimation for retail checkout

A computer-vision checkout prototype that identifies loose produce from a camera frame and estimates its weight from visual information, then calculates an estimated price using configured produce prices. Built with MobileNetV2 and Gemini 2.5 Flash.

Python
TensorFlow / Keras
MobileNetV2
Gemini 2.5 Flash
FastAPI
+3

Face Anti-Spoofing — Presentation Attack Detection cover

Computer Vision

2024

Face Anti-Spoofing — Presentation Attack Detection

Distinguishing genuine faces from presentation attacks with a lightweight CNN

A deep-learning prototype for detecting presentation attacks — printed photographs, screens and video replays — using MobileNetV3. Built to explore lightweight, on-device anti-spoofing suitable for real-time use.

Python
Deep Learning
CNN
MobileNetV3
PyTorch
+1

Marrakech Muse — Conversational Assistant for Moroccan Tourism cover

NLP
Generative AI

2024

Marrakech Muse — Conversational Assistant for Moroccan Tourism

Travel recommendations that understand what you asked for

A natural language travel assistant generating personalised recommendations for Morocco, built on LLMs and structured prompting. An earlier personal project exploring intent extraction and grounded generation for a constrained domain.

Python
NLP
LLMs
Prompt Engineering
Gemini API
+1