A local-first monitoring pipeline that turns system telemetry into calibrated anomaly signals, threshold forecasts, and actionable terminal alerts.
A terminal system monitor that collects local telemetry, scores it for anomalies with a calibrated Isolation Forest, forecasts threshold crossings, and raises alerts — without sending a single host metric anywhere.
Python · Scikit-learn · SQLite · Rich
96.2%accuracy on the synthetic benchmark
Measured on generated data — 100,000 training and 20,000 test observations with 5% injected anomalies. Not measured on real workstation traces.
CODE-VERIFIED
3.8%false-positive rate
Same synthetic benchmark, against a calibration target of 5% or below. Precision was 84.1% and recall 78.5% on the same run.
A collaborative seismic segmentation system that compares U-Net variants and packages inference behind a web and API deployment.
PyTorch · Flask · FastAPI · Docker
97.2%average accuracy — InceptionV3
The strongest of the three backbones. ResNet-50 reached 96.7% and ResNet-34 93.5% on the same task; the three-model mean is 95.8%, which is what the résumé's 95.7% figure refers to.
A multimodal misinformation platform that verifies claims from video through speech, text and retrieval — and marks generated media at source so provenance survives re-upload.
Python · Faster-Whisper · RoBERTa-MNLI · C2PA
90%+claim verification accuracy
Measured by the team during the sprint, checking extracted claims against scraped Wikipedia articles after a polarity score narrowed which statements were worth verifying.
Supervised research on exercise-form assessment from video, comparing template matching, classical ML and deep learning.
+
Show details for Undergraduate Research Assistant at European Universities in Egypt
Recorded and annotated a boxing-gesture corpus covering five actions at three execution-quality levels, tracking eight body landmarks per frame with MediaPipe.
Built a reproducible evaluation protocol — one fixed unseen test video, timestamp-aligned scoring, and the same test set applied to every approach.
Compared a $1 Recognizer, a nine-model classical sweep, and a CNN-LSTM, and reported the negative results as carefully as the positive ones.
Full-cycle functional and non-functional testing on an enterprise Test Management System.
+
Show details for QA Engineer Intern at Qestit Group
Wrote and executed UI, API and database test suites with Playwright, Postman and JMeter against a system serving a six-figure user base.
Surfaced more than twenty critical pre-release defects across the UI, API and database layers.
Identified performance bottlenecks that drove a 61% improvement in page load times, unblocking enterprise client release readiness.
Playwright · Postman · JMeter · SQL
Dell Technologies
Summer Intern — AI/ML
–
Owned the verification pipeline and the watermarking layer of a multimodal misinformation platform, presented to Dell executive leadership.
+
Show details for Summer Intern — AI/ML at Dell Technologies
Selected from a field of more than twelve thousand applicants.
Built the claim-verification pipeline: speech recognition, OCR, web retrieval, and RoBERTa-MNLI entailment over retrieved evidence.
Designed the prevention side as well — C2PA-signed provenance manifests, visible badge and QR overlays, and perceptual-hash fingerprinting for near-duplicate detection.
Presented the system and its business case to Dell executive leadership.
