All Projects
Medical AINLPGenerative AIFeatured
SHIFAA MAROC — Arabic Medical Assistant & Mental Support
An intelligent, secure, and adapted AI medical assistant tailored for Moroccan and Arabic-speaking users. Features multimodal CNN modules (EfficientNet/DenseNet for X-Ray and dermatology) and a hybrid RAG pipeline with a custom distress detection model.
Tech Stack
Backend PythonVision (PyTorch, MONAI)RAG (FAISS, BM25)Déploiement DockerFrontend Streamlit
Keywords
Backend PythonVision: PyTorch, MONAIRAG: FAISS, BM25Déploiement DockerFrontend Streamlit
Visualizations1 Charts
Deep Dive
SHIFAA MAROC is a comprehensive medical AI assistant designed specifically for Arabic-speaking and Moroccan patients and doctors.
Core Components:
- ▸Multimodal Diagnoses: Utilizes custom CNN architectures (EfficientNet-B3 for dermatology, DenseNet-121 for Chest X-Ray, ResNet-50 + MONAI for Brain MRI, and Inception-V3 for mammography density) to process medical imaging, reaching over 90% accuracy.
- ▸SHIFA-Mental Support: An integrated mental health module featuring distress detection using clinical scoring (PHQ-9) and keyword matching. Under the hood, it utilizes fine-tuned Arabic medical transformers (Jais-13B, AraBERT) to provide Cognitive Behavioral Therapy (CBT) support and crisis hotlines routing.
- ▸Hybrid RAG Pipeline: Combines FAISS (dense embeddings) and BM25 (lexical) with Reciprocal Rank Fusion (RRF) and cross-encoder reranking to answer questions across 16 medical domains in under 2 seconds.
- ▸Localization: Tailored with verified Moroccan emergency hotlines, Arab culture nuances, and a beautiful dark-mode RTL layout.