Assem ElQersh
I build intelligent solutions with AI & Deep Learning
Computer Software Engineering graduate from E-JUST specializing in computer vision, medical AI pipelines, and applied Generative AI. Experienced with Python, PyTorch, TensorFlow, FastAPI, and translating deep learning research into reliable software systems.
About Me
I am an AI / ML Engineer specializing in PyTorch, computer vision, and scalable FastAPI backend services. I graduated with a B.Sc. in Computer Software Engineering from the Egypt-Japan University of Science and Technology (E-JUST).
Most recently, I led deep learning research as Head of Deep Learning R&D at MechaML, validating novel architectures and translating findings into deployable services. My academic work spans 4 IEEE conference publications across medical imaging, CRT generative models, and TinyML.
Beyond research, I build open-source developer tooling like ghdigest on PyPI, author technical articles across LinkedIn and Medium, and actively support student leadership through IEEE.
Projects
Showing 18 intelligent systems, medical vision models, and engineering architectures.
MedFlow-V2 - AI Hospital Management System
Production-style healthcare platform combining role-based clinical records with Google MedGemma vision-language inference for chest X-ray interpretation and structured reporting.
AIOps Brain - Local-First AI Operations Assistant
Local-first infrastructure operations assistant employing Mistral function calling and n8n webhook orchestration for incident triage, contextual logging, and Slack escalation.
NLP Projects Suite
Comprehensive implementation of 8 NLP tasks covering sentiment classification, named entity recognition, topic modeling, extractive QA systems, and summarization.
Personal Assistant System
Intelligent voice-controlled personal assistant with natural language processing capabilities, task automation, and smart scheduling features.
Face Mask Detection
Real-time edge vision safety system achieving 99.2% accuracy using PyTorch and MobileNetV2. Optimized for mobile and constrained devices with inference times under 100ms.
Arabic Sign-to-Text Translation System
Real-time assistive system translating continuous Arabic Sign Language into readable text using MediaPipe 3D skeletal landmarks and temporal recurrent networks.
Object Detection Applications
Browser-based applications using TensorFlow.js for real-time object detection with multiple detection models including COCO-SSD and BlazeFace.
Rock-Paper-Scissors Gesture Game
Real-time interactive game using OpenCV and hand gesture recognition. Players compete against AI with automatic score tracking and low latency.
Color Identification Lab
Computer vision application for precise color detection and identification using advanced OpenCV color-space algorithms and clustering.
Ishihara K-Means Medical Vision Library
Medical computer vision Python package with CLI for color blindness detection using K-means clustering on Ishihara plates for automated clinical assessment.
SIFT Feature Extraction & Matching
Scale-invariant feature transform (SIFT) algorithm implementation with homography estimation and RANSAC outlier rejection for invariant object recognition.
NewsLies - Arabic Fake News Detection
Arabic misinformation classification model utilizing fine-tuned AraBERT and bidirectional LSTMs on the Arabic Fake News Dataset (AFND) with custom normalization.
Real-Time Face Detection & Landmarks
Dual-architecture face detection system with HOG detection, 68-point facial landmarks, SQLite storage, and DeepFace integration.
MIPS Microarchitecture Designs
Verilog implementations of MIPS processors covering single-cycle, multi-cycle, and pipelined architectures with hazard mitigation and simulation testbenches.
Robotics Projects Hub
Open-source robotics engineering repository containing pathfinding algorithms, PID control systems, and kinematic simulations.
Particle Collision Physics Engine
Real-time physics engine written in C using Raylib for interactive 2D particle dynamics, elastic collisions, and spatial grid hashing.
MediTrack-HMS: Hospital Management System
Desktop hospital administration system in Java featuring modular clinical record storage, appointment scheduling, and role-based data validation.
Train Ticketing & Reservation System
Web-based railway reservation system supporting passenger seat booking, CSV batch uploads for train schedules, and real-time inventory tracking.
Open Source Contributions
Published Python packages on PyPI, developer tooling, and contributions to community Generative AI projects.
RepoDigest (ghdigest v1.1.0)
Published open-source Python CLI and FastAPI package on PyPI. Parses repository ASTs, estimates LLM context token budgets, and generates structured digests for AI models.
Open Generative AI
Open-source contributor to the community generative AI project, supporting model integrations and workflows for alternative image generation tools.
Technical Skills
Core programming languages, frameworks, and tools used across production applications and research models.
Programming Languages
AI & Deep Learning
Databases & Tools
Research & IEEE Papers
Peer-reviewed scientific contributions across medical super-resolution, CRT signal autoregression, edge TinyML, and wireless physiological sensing.
A Systematic Framework for Evaluating the Diagnostic Impact of Super-Resolution Models in Medical Imaging
Evaluated 5 CNN-, GAN-, and Transformer-based super-resolution models on pediatric chest X-rays under clinical domain shift. FreqFormer achieved an AUC of 0.87 and recovered 52% of the lost diagnostic performance without retraining the downstream classifier.
Path-Aware Positional Encoding for Raster-Scan Autoregressive Image Generation in CRT Signal Space
Demonstrates that removing redundant 1D sequence positional encoding yields a 0.73 bits-per-dimension improvement on Oxford Flowers 102 at 32x32 resolution. Traversal locality across nine 2D spatial scan paths correlated with log-likelihood at r = 0.996.
Resource-constrained Dual-task Edge Intelligence for Agricultural Monitoring Using Multi-preprocessing Architectures
Engineered an autonomous ESP32 edge device combining an INT8-quantized YOLOv8s-cls leaf disease model with a compact C decision tree for soil hydration. A 14-day continuous field deployment reduced irrigation water consumption by 37.4% compared to timer baselines.
Where Does Identity Hide in Wi-Fi Sensing? Localizing Leakage Shortcuts in CSI Physiological Data
Analyzed subject identity leakage in Channel State Information (CSI) vital sign extraction. While linear probes reconstructed 100% of subject identities from raw signals, first-difference high-pass filtering and DC removal suppressed probe accuracy to the chance baseline of 0.333.
Technical Writing
Guides, deep dives, and model reviews published across Medium, LinkedIn, and IEEE student branch channels.
Scientific Publishing: Your Full Guide to Publish Your Paper
Comprehensive guide for engineering students on navigating manuscript preparation, peer review, and academic ethics.
MedGemma: Google's Open-Source Medical AI for Healthcare Innovation
Technical breakdown of Google's vision-language model for radiology and its practical deployment in clinical pipelines.
Aya Vision: Cohere's New Multilingual Vision Model
Exploration of Cohere's multilingual vision-language architecture and its performance across non-English benchmarks.
NumPyNuggets: Uncovering the Power of NumPy in Data Science
Vectorization, memory layouts, and broadcasting techniques for high-performance Python data wrangling.
Introduction to Algorithms
Foundational breakdown of algorithmic complexity, asymptotic analysis, and classic search/sort paradigms.
Demystifying Pseudo-Randomness
Linear congruential generators, statistical randomness tests, and PRNG implementations in modern software.
Experience & Community
Professional appointments, research leadership, and IEEE student community initiatives.
Head of Deep Learning R&D
- ›Directed deep learning research pipelines, validating novel model architectures and loss formulations against strict empirical baselines.
- ›Led experimental methodology, peer-reviewed publication drafting, and domain literature sweeps.
- ›Engineered deployable PyTorch prototypes, converting theoretical findings into concrete microservices with FastAPI and Docker.
- ›Organized technical reviews, cross-team knowledge transfers, and research roadmaps.
Generative AI & Python / Web Development Intern
- ›Implemented generative AI pipelines and Python backend services for enterprise automation workflows.
- ›Constructed interactive web dashboards for model diagnostics and real-time inference monitoring.
Machine Learning Intern
- ›Engineered automated feature extraction and tabular/image preprocessing pipelines.
- ›Trained and benchmarked computer vision models within a collaborative Git review environment.
Machine Learning Intern
- ›Constructed convolutional neural networks and sequence models for structured datasets.
- ›Built reproducible Python data-wrangling pipelines with NumPy and Pandas.
Data Science Intern
- ›Built predictive baseline models for time-series forecasting and multivariate tabular datasets.
IEEE Student Leadership
Directed operational finances, budgeting, and corporate sponsorships across 10+ student technical workshops and bootcamps.
Coordinated logistics and hardware tracks for JAC-ECC 2024, RoboRave 2025, and FireFighting Boat 2025 competitions.
Represented E-JUST at the IEEE Region 8 Computer Society Student & Young Professional (SYP) Congress 2024.
Promoted technical workshops and cross-border hackathons across IEEE student branches.
Coordinated regional student initiatives and leadership development across the Middle East and North Africa.
Organization
A volunteer initiative I built to bring industrial electronics AI projects under one open research hub.
IEEE IES Volunteer Initiative
This lab is an independent volunteer initiative created to combine all industrial electronics AI projects into one open research lab. Built by IEEE E-JUST Student Branch active members who are volunteers in the IEEE Industrial Electronics Society. It is not an official IEEE IES chapter or branch and is created independently in the spirit of IEEE's mission: advancing technology for humanity, aligned with IES publication tracks (TIE, TII, JESTPE, IECON).