AI / ML Engineer
Osama Sleem
Computer Vision & Generative AI — Cairo, Egypt
I build production computer-vision and generative-AI systems — from diffusion-based try-on and 3D point-cloud pipelines to bilingual speech recognition — and ship them with the MLOps to back it up.
// snapshot
- ASR infra cost reduction80–90%
- Manual aligner-prep time cut60%
- 3D staging directional F10.99
- Shipped projects & models9+
- Peer-reviewed publication1
Featured Work
Three systems, start to finish
ZAY Marketplace — AI Platform
Diffusion-based virtual try-on for realistic garment transfer, plus a visual search & product-similarity engine on deep embeddings and FAISS. Deployed as scalable FastAPI/Docker inference services with async processing and GPU acceleration, backed by a full MLOps pipeline.
3D Aligner Orthodontics Planning
End-to-end pipeline predicting cumulative and stage-wise tooth transformations. Built and annotated a 3D dental dataset via manual segmentation, then used DGCNN and PointNet++ for point-cloud feature extraction with a Transformer decoder to model sequential jaw movements.
X-ray AI Diagnostic Assistant
TensorFlow classifier for Tuberculosis, Pneumonia and Knee Osteoarthritis using a custom CNN, benchmarked against fine-tuned VGG and ResNet backbones. Integrated LangChain to power a chatbot that interprets diagnoses and communicates results interactively to patients.
More Projects
Research, experiments & competitions
Facial Attribute Manipulation — CLIPInverter + PTI
A CLIPInverter + Pivotal Tuning Inversion pipeline for text-driven facial edits with high realism and identity retention.
Symptom-to-Medical-Image Generator
Fine-tuned Stable Diffusion with LoRA to generate X-ray, CT and MRI images directly from natural-language symptom descriptions.
Room Design Image Generation
Customized Stable Diffusion via Hugging Face Diffusers to generate realistic room designs from natural-language prompts, trained on 1,000+ paired room images.
Face Photo Resolution (SRGAN)
A Super-Resolution GAN built from scratch to upscale low-resolution facial images, validated with consistent qualitative gains in clarity and sharpness.
Football Player Analysis Modeling
ML models evaluating player rating, market value and potential to support talent scouting and recruitment planning.
Turba — AI-Powered Smart Farming System
Crop recommendation model trained on soil datasets to improve reliability for small and mid-scale farms. 1st place, Samsung climate hackathon.
About
Machine learning across domains
Machine learning engineer skilled in applying AI across diverse domains such as medical imaging and real estate design. I've led a 3D aligner orthodontics planning pipeline, optimized X-ray imaging for diagnostics, and fine-tuned generative models for real estate visualization — always with an eye toward training and deploying models that hold up in production.
- Built the virtual try-on and visual search/FAISS retrieval systems, deployed via FastAPI/Docker with async, GPU-accelerated inference.
- Owns the MLOps pipeline: containerization, deployment, versioning, monitoring, CI/CD.
- Built bilingual (Arabic–English) ASR models optimized for CPU-only inference at ~25% WER, cutting infra cost 80–90% vs. GPU deployment.
- Owned the full ASR pipeline: audio preprocessing, custom tokenization, training, evaluation, production integration.
- Teach Deep Learning, Computer Vision, and 3D Modeling; mentor students shipping real projects with PyTorch, TensorFlow, and OpenCV.
- Fine-tuned Stable Diffusion models for automated real estate visualization.
- Assessed AI tools across business units to guide technology adoption.
- Supervised/unsupervised learning, object detection, classification, and GANs — 2nd place among AI interns.
- Ran end-to-end ML workflows and presented results internally.
GPA 3.7/4.0 · Top 5 Students, Artificial Intelligence college
Computer Vision & Generative AI
3D & Deep Learning
MLOps & Engineering
Libraries & Languages
Comparative study of Custom CNN, VGG-16 and ResNet for diagnosing tuberculosis, pneumonia and osteoarthritis — up to 96% accuracy with VGG-16.
Contact
Let's talk about computer vision, generative AI, or shipping ML to production
Open to roles and collaborations in computer vision, generative AI, and MLOps. Based in Cairo, Egypt — available remote or hybrid.