I am Mohamed Sheref, an undergraduate in Systems & Biomedical Engineering at Cairo University. Co-founder & AI Lead at OShift. Working on physics-informed neural networks for biological systems, deep reinforcement learning, medical imaging, and autonomous agentic software.
My work centers on the intersection of biological systems, continuous mathematics, and high-performance computing. From designing desktop DICOM/NIfTI volume reconstructors with real-time oblique slicing to formulating discrete Markov Decision Processes for pharmaceutical logistics, I build software grounded in physical and data-driven rigor.
In production, I co-founded OShift to bring multi-step agentic workflows to enterprise marketing, interned in clinical biomedical engineering at Children's Cancer Hospital Egypt 57357, and lead technical AI curricula for 60+ engineers as Vice Head at IEEE EMBS CUSB.
Python · PyTorch · TypeScript · Next.js 16 · FastAPI · Supabase · DICOM / NIfTI · PyVista 3D · PostgreSQL · Inngest · Signal DSP · PINNs
Nine projects and research implementations across medical imaging, reinforcement learning, agentic AI, and the web.
01
Co-founder build: autonomous marketing operating system with persistent memory, multi-step scheduling, and automated tool-use loops across web and social platforms.
Next.js 16 · FastAPI · Supabase · Inngest
02
Automated learning platform converting uploaded syllabi into flashcards, Suno musical songs, STEM diagrams, and auto-graded tests.
Gemini API · Next.js · Supabase
03
Pharmaceutical inventory optimization modeled as a discrete MDP. PPO policy cuts stock spoilage by ~9% vs standard heuristics.
PPO Policy · MDP Formulation · StockOut Game
04
Match outcome prediction pipeline for post-match fault analysis: 87.4% win accuracy and 52.0% score diff R² on World Cups with SHAP attribution.
icSPORTS '26 · SHAP Attribution · Neural Nets
05
Desktop medical volume workstation supporting multi-planar reconstruction (axial, coronal, sagittal), oblique slicing, and PyVista 3D mesh rendering.
PyQt5 · PyVista 3D · DICOM / NIfTI
06
Physics-informed neural networks and Laplace Neural Operators modeling nonlinear differentiation kinetics under ODE constraints with ~1 ms inference.
PINNs · Laplace Operators · PyTorch
07
Benchmarking classical Butterworth, Notch, and adaptive LMS/RLS filter topologies across ECG signals while preserving clinical QRS morphology.
Butterworth · Notch · Adaptive LMS
08
Modular medical image segmentation repository with custom U-Net architectures, standardized training loops, and Dice/IoU metric evaluation.
Custom U-Net · Dice Loss · PyTorch
09
Personal research archive rendered as an interactive force-directed graph where technical essays exist as interconnected nodes.
Next.js · MDX · D3 Force Graph
Credentials across machine learning, cloud engineering, clinical biomedical systems, and web development.
Deep neural networks, CNNs, Transformers & sequence models, hyperparameter tuning, and optimization — Andrew Ng's specialization.
Verify Certificate ↗Supervised learning, deep neural networks, decision trees, and unsupervised learning — Andrew Ng's specialization.
Verify Certificate ↗Cloud ML pipelines, AWS SageMaker, computer vision, and generative AI foundations.
Verify Certificate ↗Production React, modern state architecture, and API systems engineering.
Scholar RecordClinical diagnostic medical equipment, MRI/CT calibration, and radiotherapy protocols.
Clinical Trainee RecordLevel Three web development: full-stack web architecture, algorithms, and software engineering.
Verify Certificate ↗The full record: roles, research, clinical training, awards, and education.
Open to internships, research collaborations, and high-impact engineering challenges.
me@sheref.dev ↗