Every engineer has an origin story. Mine runs through biology, medicine, and mathematics — and ends in code.
It started with a simple question — how can machines learn to see what doctors see? That question pulled me from circuits and biology into the world of algorithms.
I taught myself Python, then machine learning, then deep learning — reimplementing published papers line by line until the architectures stopped being diagrams and started being code I could reason about.
Biomedical engineering gave me the domain; AI gave me the tools. I work at their intersection — volumetric segmentation, histopathology, clinical data — and the same intuitions carry over anywhere structure hides in messy signals, from medical scans to football tracking data.
Today I build AI systems end to end: research-grade models in PyTorch on one side, production RAG pipelines with FastAPI, Docker, and vector databases on the other — turning research ideas into software people can actually use.
The tools I reach for daily — from research-grade training loops to deployed inference services.
Internships and applied training — each step added a new dimension to how I build.
FlyRank AI
Built reproducible ML workflows in Python and scikit-learn over large-scale search intelligence datasets — covering feature engineering, model training, and evaluation — and analyzed search performance signals to produce data-driven content optimization recommendations delivered to the product team.
Digital Egypt Pioneers Initiative
Completed an intensive applied data science program in Python, SQL, statistics, and machine learning, delivering end-to-end predictive modeling projects on real-world datasets from data cleaning and feature engineering through model selection and evaluation.
NAID
Worked on innovative technology solutions for individuals with disabilities, focusing on biomedical applications.
Summit Technology Solutions
Gained hands-on experience working with large-scale Oracle data systems using SQL.
Selected work across deep learning research, medical imaging, and production AI systems — each built end-to-end, from data to deployment.
Reimplemented the CVPR 2016 paper in modern PyTorch, replacing the original AlexNet/Caffe stack with a ResNet50 + two-stage Person LSTM & Group BiLSTM pipeline that models both individual player actions and team-level dynamics from volleyball footage.
+9.95% over the published CVPR result · surpassed all 8 baselines
Engineered a Low-Rank Adaptation framework from scratch to transfer the deep spatial features of UNETR++ — a 42M-parameter Vision Transformer — into a data-scarce 3D segmentation domain without full fine-tuning.
0.94 Mean Dice vs. 0.95 full fine-tune · 99.5% fewer trainable params (135k)
An asynchronous Retrieval-Augmented Generation backend for bilingual (Arabic/English) document intelligence: ingestion, semantic search, and grounded question answering with multi-provider vector database and LLM support.
Bilingual AR/EN ingestion · OpenAI, Cohere & Ollama providers
An automated ML pipeline that bypasses deep learning black boxes by extracting morphological, color, and textural features from histopathologic scans to identify metastatic tissue — every decision traceable to a named feature.
0.9542 ROC AUC · fully interpretable feature set
A hierarchical Transformer that maps multi-agent tracking data from the FIFA World Cup 2022 into tactical embeddings for similarity retrieval over plays, trained with supervised contrastive learning on weakly-derived labels. A research collaboration with teaching assistants at a German university.
Research collaboration · FIFA World Cup 2022 tracking data
An engineering education built on mathematics, medicine, and computer science — with a focus on deep learning, computer vision, and biomedical systems.
Cairo University — Faculty of Engineering · Giza, Egypt
2022 — 2027 (expected)
These are the frontiers I’m working toward — the goals that guide what I learn and build today.
Clinically trustworthy models for diagnosis and monitoring.
Perception systems that understand medical and natural imagery.
Attention-based architectures across vision and language.
Grounded, domain-adapted LLMs for specialized knowledge.
Doing more with less compute, data, and energy.
Compressing networks without losing what matters.
Reliable pipelines from experiment to production.
Formal milestones along a self-driven learning path.
A concise summary of my education, experience, and projects — ready for your team.
Amr Ahmed — Resume
PDF · 1 page · Updated 2026
Whether it’s a role, a research collaboration, or just an idea worth discussing — my inbox is open.
