I build production-grade intelligent systems that combine numerical optimization with domain physics, developing models that are both data-efficient and physically consistent. additionally My work spans performance optimization, robust modeling, inference acceleration, and deploying AI solutions in resource-constrained environments.
π Cairo, Egypt | π§ hadywaliedkamel@gmail.com
The Problem: Large transformer models are prohibitively expensive for deployment in production environments with latency and resource constraints.
The Solution: End-to-end compression pipeline combining knowledge distillation and quantization. Implemented custom training loop with PyTorch Lightning, integrated W&B for experiment tracking, and built production-ready inference engine.
Results: 75% model size reduction β’ 3x faster inference β’ <2% accuracy loss β’ Deployed with Qt-based demo application
Stack: PyTorch Lightning, Transformers, W&B, Quantization, Model Distillation
The Problem: Traditional EDA optimization approaches ignore underlying physical constraints, leading to unrealistic solutions and poor generalization.
The Solution: Hybrid optimization system combining gm/ID methodology-based methods with physics-informed constraints. Built custom computational engine leveraging NumPy/Numba/Sympy for performance-critical operations and integrated semiconductor-specific physical models.
Impact: multiple folds reduction in solution iteration time β’ Deployed to production serving industrial applications
Stack: Python, C++, NumPy, Numba, Optimization Algorithms, AWS
cGrad - ML Fundamentals from Scratch
The Challenge: Understanding automatic differentiation and backpropagation at the implementation level.
The Solution: Lightweight autograd engine and neural network library built in modern C++17βno framework dependencies. Implements core ML primitives: computational graphs, reverse-mode autodiff, gradient descent optimizers, and basic neural network layers.
Purpose: Deep dive into ML fundamentals, performance-critical C++ design, and educational resource for learning autodiff mechanics.
Stack: Modern C++17, CMake, Template Metaprogramming
- Building a Simple Modern RAG Application with Asyncio and Chainlit
- From Documents to Dialogue: A step-by-step RAG Journey
ML/AI Core: PyTorch (Lightning) β’ TensorFlow β’ scikit-learn β’ Model Compression β’ Knowledge Distillation
Performance Engineering: C++ β’ Rust β’ Python optimization β’ Numba β’ CUDA basics β’ Memory profiling
MLOps & Deployment: Docker β’ AWS (EC2, S3, CI/CD) β’ Model serving β’ Experiment tracking (W&B) β’ MLFlow β’ ETL (Pandas)
Scientific Computing: NumPy β’ SciPy β’ Statistical modeling β’ Optimization algorithms β’ DSP
- Physics-Informed Neural Networks (PINNs) for solving differential equations and inverse problems
- Model optimization techniques: Pruning, quantization, distillation for edge deployment
- High-performance ML inference: Exploring Rust and C++ for production ML systems
- Hybrid approaches: Combining classical optimization with deep learning
B.S.E. Electronics & Electrical Communications Engineering β’ Cairo University β’ 2021
Relevant Coursework: Linear Algebra, Calculus (ODE/PDE), Classical & Deep ML, DSP, Statistical Methods
Professional Certifications:
- Deep Learning Specialization (Coursera - deeplearning.ai)
- NLP Specialization (Coursera - deeplearning.ai)
- Google Certified Associate Android Developer (2020-2023)
I'm actively seeking roles in:
- ML Engineering: Model development, optimization, and production deployment
- Applied AI Research: Physics-Informed ML, model compression, efficient inference
- ML Systems Engineering: High-performance inference engines, C++/Python integration
- Research Scientist positions: PIML, hybrid physics-ML approaches, scientific ML
Best way to reach me: LinkedIn or email