AI & Machine Learning Engineer · Cairo, Egypt

I build ML systems end to end.

From data collection and honest evaluation to web apps people can actually use. My graduation project fuses text, audio and image models into one dog-health assistant.

See the projects Pawlytics case study Download CV
Public projects
21
Live demos
6
Modalities in one product
3
Conditions classified
20

Featured · Graduation project

Pawlytics, a multimodal AI pet-health assistant

Describe the symptoms, upload a photo, record a sound. Three deep-learning models run behind an Express and Flask app, and a rule-based step merges them into one report with likely conditions, an urgency level and next steps.

Text

Bio_ClinicalBERT + 256-unit head · 20 conditions

63.8%

Held-out test accuracy, split by dog. Symptom text was generated from diagnoses with rules; a noisier, more realistic run scored 42.2%.

Audio

Frozen YAMNet embeddings + MLP · 8 sound classes

71.9%

Test accuracy with macro-F1 0.71 on 185 held-out clips. Weakest class: normal playing (recall 0.33).

Image

EfficientNet-B0, frozen backbone + trained head · 9 classes

73.2%

Test accuracy on 1,752 held-out images merged from 20+ public datasets. The 94.1% cross-validation figure uses gradual unfreezing, a different setup.

Pawlytics multi-modal analysis form with symptom text, photo and audio upload
  • Any mix of inputs. Symptom text with breed, age and sex, a photo of the skin, eyes or nose, or a short recording. The report merges whatever was supplied.
  • An inspectable final step. Fusion uses thresholds and keyword rules instead of another learned model, so every recommendation can be traced.
  • Honest evaluation. Text split by dog to avoid leakage, held-out test sets, and the caveats written down in the model cards.
  • A real product around the models. Accounts, dog profiles, medical history, a dashboard of past analyses and rate-limited, validated endpoints.
View the repository Read the model cards

Selected work

All projects

Every project has its code on GitHub with a short write-up. Six also run live in the browser.

About

Trustworthy ML, not just big numbers

I collect and clean data, train and honestly evaluate models, and wrap them in web apps. I care about the parts that make ML trustworthy: leak-free splits, class imbalance, reporting held-out results, and being upfront about what a model can't do.

These projects come from my computer science studies and my graduation project. Right now I'm continuing to develop Pawlytics, exploring vector search and memory-augmented models, and learning React and Docker.

Toolbox

Languages
Python · JavaScript · C++ · SQL · Bash
ML & AI
PyTorch · TensorFlow · Keras · Hugging Face · scikit-learn · XGBoost · OpenCV
Data
pandas · NumPy · Jupyter · Kaggle
Web
Node.js · Express · Flask · MongoDB · SQLite · HTML · CSS · p5.js
Tools
Git · GitHub · Docker (basics)

Let's talk

Open to internships, collaboration and project opportunities.

ziadelhussein@gmail.com LinkedIn GitHub Download CV