Omar Ashraf Mohammed — home

Projects

Selected engineering work

Five pieces of work, chosen for what they show rather than for how many there are. Systems, models, data and deployment — each written up with the decisions, the evaluation context and the parts that did not work.

All work

PERSONALML SYSTEMS2025

SysMon AI

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.

CODE-VERIFIED

Read case study

INTERNSHIP CASE STUDY2025

AI Facies Classification

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.

AUTHOR-CLARIFIED

Read case study

INTERNSHIP CASE STUDY2025

SpotAI

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.

AUTHOR-CLARIFIED

Read case study

A relational data platform for examining what food-hygiene ratings reveal — and conceal — across seven authorities.

MySQL · Node.js · Express · EJS

21,209 establishments modelled

Food Standards Agency FHRS records across seven local authorities, loaded and validated into a normalised MySQL schema.

REPORT-VERIFIED

Read case study

INTERNSHIP CASE STUDY2026

Orange Promotional Gift Fulfilment

A production-style promotional fulfilment engine designed to grant a gift exactly once under concurrent requests.

Java 21 · Spring Boot · PostgreSQL

Five concurrent calls in the exactly-once test

The integration test issues five simultaneous fulfilment requests for one eligible user and asserts exactly one grant, with the rest returning a consistent replay.

CODE-VERIFIED

Read case study

ML systems


Applications & backend


Elsewhere