# Assem ElQersh — AI / ML Systems Engineer & Researcher

> 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 orchestration. Author of the ghdigest CLI on PyPI and creator of the MedFlow-V2 clinical AI system.

## Core Systems & Repositories

- [MedFlow-V2: Hospital Management & Clinical AI System](https://github.com/Assem-ElQersh/MedFlow-V2): Production-style clinical AI platform integrating Google MedGemma vision-language inference for automated chest X-ray interpretation with FastAPI, React, and Docker.
- [AIOps Brain: Local-First AI Operations Assistant](https://github.com/Assem-ElQersh/ai-ops-brain): Infrastructure operations assistant utilizing Mistral tool calling, local LLMs, and n8n webhook orchestration for incident triage.
- [NLP Projects Suite](https://github.com/Assem-ElQersh/NLP-Projects): Modular PyTorch and Hugging Face suite implementing 8 NLP tasks including extractive QA, sentiment classification, and named entity recognition.
- [RepoDigest (ghdigest v1.1.0 on PyPI)](https://pypi.org/project/ghdigest/): Published Python CLI and FastAPI package for AST parsing, token budget estimation, and repository serialization for LLMs.
- [Open Generative AI](https://github.com/anil-matcha/open-generative-ai): Open-source contribution for alternative generative AI model workflows and interface engineering.

## Peer-Reviewed IEEE Research Publications

- [Diagnostic Impact of Super-Resolution Models in Medical Imaging (IEEE JAC-ECC 2025)](https://ieeexplore.ieee.org/document/11417605): Evaluated 5 CNN, GAN, and Transformer models under clinical domain shift. FreqFormer achieved ROC-AUC = 0.87 and recovered 52% of lost diagnostic performance without classifier retraining. DOI: 10.1109/JAC-ECC.2025.11417605.
- [Path-Aware Positional Encoding for Raster-Scan Autoregressive Image Generation (IEEE NILES 2026)](https://www.assemelqersh.com/#research): Removed redundant 1D positional encoding to achieve a 0.73 bits-per-dimension (BPD) density gain on Oxford Flowers 102; spatial scan-path locality correlated with log-likelihood at r = 0.996.
- [Dual-Task Edge Intelligence for Agricultural Monitoring (IEEE JAC-ECC 2026)](https://www.assemelqersh.com/#research): Autonomous ESP32 microcontroller system combining INT8-quantized YOLOv8s-cls and embedded C decision tree, reducing field irrigation water by 37.4%.
- [Identity Leakage in Wi-Fi CSI Physiological Sensing (IEEE JAC-ECC 2026)](https://www.assemelqersh.com/#research): Investigated biometric privacy leakage in RF Channel State Information; suppressed 100% linear probe identity leakage down to 0.333 chance baseline via high-pass filtering.

## Technical Writing & Articles

- [Scientific Publishing: Your Full Guide to Publish Your Paper](https://edu.ieee.org/eg-ejust/scientific-publishing/): Guide on manuscript formulation, peer review navigation, and publication ethics for engineering researchers.
- [MedGemma: Google's Open-Source Medical AI for Healthcare Innovation](https://www.linkedin.com/pulse/medgemma-googles-open-source-medical-ai-healthcare-assem-elqersh-ohajf): Technical breakdown of multimodal vision-language architectures in clinical environments.
- [Aya Vision: Cohere's Multilingual Vision Model](https://www.linkedin.com/pulse/aya-vision-coheres-new-multilingual-model-assem-elqersh-l1mwf): Analysis of multilingual vision representation learning across non-English benchmarks.
- [NumPyNuggets: Uncovering the Power of NumPy in Data Science](https://medium.com/@assemelqersh/numpynuggets-uncovering-the-power-of-numpy-in-data-science-610409876880): Deep dive on vectorized execution, stride memory layouts, and broadcasting paradigms.
- [Introduction to Algorithms](https://medium.com/@assemelqersh/introduction-to-algorithms-44aa4ea251b9): Asymptotic complexity, divide-and-conquer, and algorithmic problem-solving.
- [Demystifying Pseudo-Randomness](https://medium.com/@assemelqersh/demystifying-pseudo-randomness-6398668b65e4): Linear congruential generators, statistical randomness tests, and PRNG software implementations.

## Dossier & Verification Endpoints

- [Comprehensive Technical Dossier (Markdown)](https://www.assemelqersh.com/llms-full.txt): Complete, zero-slop technical documentation containing verified benchmarks, architecture diagrams, and career history.
- [Curriculum Vitae (PDF)](https://www.assemelqersh.com/Assem%20ElQersh%20_%20CV.pdf): Two-page verified engineering resume.
- [Official Website](https://www.assemelqersh.com/): Interactive developer portfolio.