# Assem ElQersh — Comprehensive Engineering & Research Dossier

> AI / ML Engineer and Head of Deep Learning R&D at MechaML. 4x IEEE published author specializing in computer vision, clinical AI pipelines, edge TinyML, and LLM systems. Software engineering graduate from the Egypt-Japan University of Science and Technology (E-JUST).

---

## 1. Professional Background & Core Roles

### Head of Deep Learning R&D — MechaML
- **Period:** Sep 2025 – Present
- **Domain:** Deep Learning Research & Architecture Validation
- **Highlights:**
  - Led deep learning architecture design and experimental validation across clinical imaging and generative vision projects.
  - Converted empirical research breakthroughs into deployable, low-latency microservices with FastAPI and Docker.
  - Mentored junior engineers on research methodology, reproducible evaluation pipelines, and PyTorch model optimization.

### Computer Vision Intern — Information Technology Institute (ITI)
- **Period:** Aug 2024 – Sep 2024
- **Domain:** Industrial & Medical Computer Vision
- **Highlights:**
  - Engineered automated feature extraction, spatial alignment, and tabular/image preprocessing pipelines.
  - Benchmarked state-of-the-art vision architectures within a collaborative Git review environment.

### Machine Learning Intern — Mindset Training
- **Period:** Aug 2024 – Sep 2024
- **Domain:** Deep Learning & Structured Data
- **Highlights:**
  - Constructed convolutional neural networks and sequence models for structured datasets.
  - Built reproducible Python data-wrangling pipelines with NumPy and Pandas.

### Data Science Intern — Creativa Hub Alexandria
- **Period:** Aug 2024
- **Domain:** Predictive Modeling & Analytics
- **Highlights:**
  - Built predictive baseline models for time-series forecasting and multivariate tabular datasets.

---

## 2. Academic Background

- **Institution:** Egypt-Japan University of Science and Technology (E-JUST)
- **Degree:** B.Sc. in Computer Software Engineering
- **Core Coursework:** Deep Learning, Computer Vision, Robotics, Cryptography, Parallel Computing & Distributed Systems, Data Structures & Algorithms, Numerical Analysis, Computer Architecture, Operating Systems.

---

## 3. Peer-Reviewed IEEE Research Publications

### 1. A Systematic Framework for Evaluating the Diagnostic Impact of Super-Resolution Models in Medical Imaging
- **Venue:** IEEE JAC-ECC 2025 (8th International Japan-Africa Conference on Electronics, Communications, and Computations)
- **Status:** Published
- **DOI:** [10.1109/JAC-ECC.2025.11417605](https://doi.org/10.1109/JAC-ECC.2025.11417605)
- **IEEE Xplore:** [Document 11417605](https://ieeexplore.ieee.org/document/11417605)
- **Quantitative Metrics:**
  - Target Dataset: Pediatric Chest X-ray Super-Resolution under clinical domain shift
  - Top Model: FreqFormer
  - Downstream Diagnostic ROC-AUC: 0.87
  - Diagnostic Gap Recovery: 52% performance restoration without classifier retraining
- **Abstract:** 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.
- **BibTeX:**
```bibtex
@inproceedings{elqersh2025systematic,
  title={A Systematic Framework for Evaluating the Diagnostic Impact of Super-Resolution Models in Medical Imaging},
  author={ElQersh, Assem and others},
  booktitle={2025 8th International Japan-Africa Conference on Electronics, Communications, and Computations (JAC-ECC)},
  pages={1--6},
  year={2025},
  organization={IEEE},
  doi={10.1109/JAC-ECC.2025.11417605}
}
```

### 2. Path-Aware Positional Encoding for Raster-Scan Autoregressive Image Generation in CRT Signal Space
- **Venue:** IEEE NILES 2026 (8th Novel Intelligent and Leading Emerging Sciences Conference)
- **Status:** Accepted, forthcoming
- **Quantitative Metrics:**
  - Density Improvement: 0.73 bits-per-dimension (BPD) reduction on Oxford Flowers 102
  - Locality Correlation: r = 0.996 between spatial scan-path locality and log-likelihood
  - Traversal Paths Evaluated: 9 distinct 2D scanning paths
- **Abstract:** 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.
- **BibTeX:**
```bibtex
@inproceedings{elqersh2026pathaware,
  title={Path-Aware Positional Encoding for Raster-Scan Autoregressive Image Generation in CRT Signal Space},
  author={ElQersh, Assem and others},
  booktitle={2026 8th Novel Intelligent and Leading Emerging Sciences Conference (NILES)},
  year={2026},
  organization={IEEE}
}
```

### 3. Resource-constrained Dual-task Edge Intelligence for Agricultural Monitoring Using Multi-preprocessing Architectures
- **Venue:** IEEE JAC-ECC 2026 (9th International Japan-Africa Conference on Electronics, Communications, and Computations)
- **Status:** Accepted, forthcoming
- **Quantitative Metrics:**
  - Edge Microcontroller: ESP32 (offline, battery-constrained)
  - Vision Architecture: INT8-quantized YOLOv8s-cls
  - Continuous Field Validation: 14 days
  - Water Conservation: 37.4% reduction in irrigation volume versus timer baselines
- **Abstract:** 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.
- **BibTeX:**
```bibtex
@inproceedings{elqersh2026edgeintel,
  title={Resource-constrained Dual-task Edge Intelligence for Agricultural Monitoring Using Multi-preprocessing Architectures},
  author={ElQersh, Assem and others},
  booktitle={2026 9th International Japan-Africa Conference on Electronics, Communications, and Computations (JAC-ECC)},
  year={2026},
  organization={IEEE}
}
```

### 4. Where Does Identity Hide in Wi-Fi Sensing? Localizing Leakage Shortcuts in CSI Physiological Data
- **Venue:** IEEE JAC-ECC 2026 (9th International Japan-Africa Conference on Electronics, Communications, and Computations)
- **Status:** Accepted, forthcoming
- **Quantitative Metrics:**
  - Raw Identification Leakage: 100% subject identity reconstruction by linear probes
  - Sanitized Probe Accuracy: 0.333 (exact random chance baseline)
  - Signal Domain: Wi-Fi Channel State Information (CSI) RF vital signs
  - Remediation Technique: First-difference high-pass filtering + DC component removal
- **Abstract:** 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.
- **BibTeX:**
```bibtex
@inproceedings{elqersh2026wifisensing,
  title={Where Does Identity Hide in Wi-Fi Sensing? Localizing Leakage Shortcuts in CSI Physiological Data},
  author={ElQersh, Assem and others},
  booktitle={2026 9th International Japan-Africa Conference on Electronics, Communications, and Computations (JAC-ECC)},
  year={2026},
  organization={IEEE}
}
```

---

## 4. Open-Source Contributions & Packages

### 1. RepoDigest (ghdigest v1.1.0 on PyPI)
- **Package URL:** [https://pypi.org/project/ghdigest/](https://pypi.org/project/ghdigest/)
- **Repository:** [https://github.com/Assem-ElQersh/repodigest](https://github.com/Assem-ElQersh/repodigest)
- **Role:** Author & Maintainer
- **Tech Stack:** Python, FastAPI, AST Parsing, PyPI, CLI, Asyncio
- **Description:** Published open-source Python CLI and FastAPI package on PyPI. Parses repository Abstract Syntax Trees (ASTs), estimates LLM context token budgets, and generates structured digests for AI models.

### 2. Open Generative AI
- **Repository:** [https://github.com/anil-matcha/open-generative-ai](https://github.com/anil-matcha/open-generative-ai)
- **Role:** Open Source Contributor
- **Tech Stack:** React, Generative AI, Web Development
- **Description:** Open-source contributor to community generative AI tooling, supporting model integrations and alternative image synthesis workflows.

---

## 5. Complete Engineering Projects (18 Systems)

### AI & Deep Learning Systems
1. **MedFlow-V2 - AI Hospital Management System:** Healthcare platform combining role-based clinical records with Google MedGemma vision-language inference for chest X-ray interpretation and structured reporting. [Tech: Python, FastAPI, Google MedGemma, React, MongoDB, Docker]. GitHub: `Assem-ElQersh/MedFlow-V2`.
2. **AIOps Brain - Local-First AI Operations Assistant:** Infrastructure assistant employing Mistral function calling and n8n webhook orchestration for incident triage, logging, and Slack escalation. [Tech: Python, Mistral AI, Tool Calling, n8n, Slack API]. GitHub: `Assem-ElQersh/ai-ops-brain`.
3. **Personal Assistant System:** Voice-controlled personal assistant with natural language processing capabilities, task automation, and smart scheduling. [Tech: Python, NLP, Speech Recognition, AI, Automation]. GitHub: `Assem-ElQersh/Personal-Assistant`.

### Natural Language Processing (NLP)
4. **NLP Projects Suite:** Implementation of 8 distinct NLP tasks covering sentiment classification, named entity recognition, topic modeling, extractive QA systems, and summarization. [Tech: Python, BERT, Transformers, NLP, PyTorch]. GitHub: `Assem-ElQersh/NLP-Projects`.
5. **NewsLies - Arabic Fake News Detection:** Misinformation classification utilizing fine-tuned AraBERT and bidirectional LSTMs on the Arabic Fake News Dataset (AFND). [Tech: Python, TensorFlow, AraBERT, LSTM, Arabic NLP]. GitHub: `Assem-ElQersh/NewsLies`.

### Computer Vision & Edge AI
6. **Face Mask Detection:** Real-time edge vision safety system achieving 99.2% accuracy using PyTorch and MobileNetV2 with sub-100ms inference on constrained hardware. [Tech: PyTorch, MobileNetV2, OpenCV, Edge Optimization]. GitHub: `Assem-ElQersh/Face-Mask-Detection`.
7. **Arabic Sign-to-Text Translation System:** Real-time assistive system translating continuous Arabic Sign Language into text using MediaPipe 3D skeletal landmarks and recurrent LSTM networks. [Tech: Python, OpenCV, MediaPipe, TensorFlow, LSTM]. GitHub: `Assem-ElQersh/Sign-to-Text-Translation-System-Using-Arabic-Sign-Language`.
8. **Object Detection Applications:** Browser-based client applications using TensorFlow.js for real-time multi-model object detection (COCO-SSD, BlazeFace). [Tech: TensorFlow.js, React, JavaScript, Computer Vision]. GitHub: `Assem-ElQersh/Object-Detection-App-2`.
9. **Rock-Paper-Scissors Gesture Game:** Real-time interactive game using OpenCV and hand gesture recognition with low latency. [Tech: Python, OpenCV, cvzone, Hand Tracking]. GitHub: `Assem-ElQersh/Rock-Paper-Scissors-Game`.
10. **Color Identification Lab:** Computer vision application for precise color detection and identification using OpenCV color-space transformations and clustering. [Tech: Python, OpenCV, Machine Learning, Computer Vision]. GitHub: `Assem-ElQersh/Color-Identification-Lab`.
11. **Ishihara K-Means Medical Vision Library:** Medical computer vision Python package with CLI for color blindness detection using K-means clustering on Ishihara plates. [Tech: Python, Medical AI, Computer Vision, K-means Clustering]. GitHub: `Assem-ElQersh/Ishihara-Kmeans`.
12. **SIFT Feature Extraction & Matching:** Scale-invariant feature transform implementation with homography estimation and RANSAC outlier rejection. [Tech: Python, OpenCV, SIFT, Computer Vision, RANSAC]. GitHub: `Assem-ElQersh/SIFT-Feature-Extraction-and-Matching`.
13. **Real-Time Face Detection & Landmarks:** Dual-architecture face detection system with HOG detection, 68-point facial landmarks, SQLite storage, and DeepFace integration. [Tech: Python, OpenCV, DeepFace, SQLite]. GitHub: `Assem-ElQersh/Real-time-Face-Detection`.

### Hardware, Architecture & Core Systems
14. **MIPS Microarchitecture Designs:** Verilog implementations of MIPS processors covering single-cycle, multi-cycle, and pipelined architectures with hazard mitigation and simulation testbenches. [Tech: Verilog HDL, ModelSim, Computer Architecture]. GitHub: `Assem-ElQersh/MIPS-Processor-Designs`.
15. **Robotics Projects Hub:** Robotics engineering repository containing pathfinding algorithms, PID control systems, and kinematic simulations. [Tech: Python, Robotics, Algorithms, Control Systems]. GitHub: `Assem-ElQersh/RoboticsProjectsHub`.
16. **Particle Collision Physics Engine:** Real-time physics engine in C using Raylib for 2D particle dynamics, elastic collisions, and spatial grid hashing. [Tech: C, Raylib, Physics Simulation, Real-time Rendering]. GitHub: `Assem-ElQersh/Particle-Collision-Simulation`.
17. **MediTrack-HMS: Hospital Management System:** Desktop hospital administration software in Java with modular clinical record storage and appointment scheduling. [Tech: Java, GUI, Database Management, Healthcare Systems]. GitHub: `Assem-ElQersh/MediTrack-HMS`.
18. **Train Ticketing & Reservation System:** Railway reservation system supporting seat booking, CSV batch uploads for train schedules, and inventory tracking. [Tech: JavaScript, PHP, MySQL, CSS]. GitHub: `Assem-ElQersh/Train-Ticketing-System`.

---

## 6. Technical Writing & Publications

1. **Scientific Publishing: Your Full Guide to Publish Your Paper** (IEEE E-JUST SB, April 2025): Detailed methodology on scientific manuscript preparation, experimental validation, peer review response strategies, and academic integrity.
2. **MedGemma: Google's Open-Source Medical AI for Healthcare Innovation** (LinkedIn, June 2025): Architecture and clinical utility breakdown of Google's vision-language model for automated radiology reporting.
3. **Aya Vision: Cohere's New Multilingual Vision Model** (LinkedIn, March 2025): Technical examination of multilingual vision-language modeling across 23 languages.
4. **NumPyNuggets: Uncovering the Power of NumPy in Data Science** (Medium, Oct 2023): In-depth guide on SIMD vectorization, memory strides, broadcasting rules, and memory layout optimization.
5. **Introduction to Algorithms** (Medium, Sep 2023): Complexity classes, asymptotic notation, and algorithmic design patterns.
6. **Demystifying Pseudo-Randomness** (Medium, Aug 2024): Mathematical underpinnings of Linear Congruential Generators (LCGs), PRNG security, and randomness testing.

---

## 7. Technical Skills & Tooling Taxonomy

- **Programming Languages:** Python, C++, C, Java, TypeScript, JavaScript, Verilog HDL
- **AI & Deep Learning:** PyTorch, TensorFlow, OpenCV, NumPy, Pandas, Scikit-Learn, Hugging Face Transformers, MediaPipe, DeepFace, YOLOv8
- **Backend & Cloud Services:** FastAPI, Docker, MongoDB, SQLite, Git, Linux / Bash, REST APIs, Microservices
- **Frontend & Visualization:** React, Next.js, Tailwind CSS, Three.js, GSAP, HTML5 / CSS3

---

## 8. Leadership & Community Initiatives

- **Treasurer — IEEE E-JUST Computer Society Student Branch (Oct 2024 – Oct 2025):** Managed finances and sponsorship budgets across 10+ bootcamps, JAC-ECC 2024, RoboRave 2025, and FireFighting Boat 2025 competitions.
- **IEEE IES Volunteer Initiative:** Open research hub advancing industrial electronics AI benchmarks under the MIT License (`github.com/IEEE-IES-Industrial-AI-Lab`).
- **Ambassador Roles:** IEEE CS SYP TechX Ambassador (2025–2026) and IEEE Region 8 MENA SYP Ambassador (2025).

---

## 9. Contact & Official Verified Channels

- **Portfolio:** [https://www.assemelqersh.com/](https://www.assemelqersh.com/)
- **GitHub:** [https://github.com/Assem-ElQersh](https://github.com/Assem-ElQersh)
- **LinkedIn:** [https://www.linkedin.com/in/assemelqersh/](https://www.linkedin.com/in/assemelqersh/)
- **X / Twitter:** [https://x.com/AssemElQersh](https://x.com/AssemElQersh)
- **Medium:** [https://medium.com/@assemelqersh](https://medium.com/@assemelqersh)
- **PyPI Package:** [https://pypi.org/project/ghdigest/](https://pypi.org/project/ghdigest/)
- **Email:** `assem.elqersh@gmail.com`