Hello, I'm
Mahmoud Osama
|
Backend-focused AI engineer building production systems with Python, FastAPI, and Django. I ship ML pipelines, LLM-powered applications, and security tools end-to-end.
Hello, I'm
Backend-focused AI engineer building production systems with Python, FastAPI, and Django. I ship ML pipelines, LLM-powered applications, and security tools end-to-end.
Backend-focused AI engineer working primarily with Python, FastAPI, and Django. My experience spans ML pipelines, LLM automation, and security applications—phishing detection and red teaming in particular.
With a freelance background, I've shipped production systems end-to-end: from client requirements through architecture decisions to deployment. I build AI applications that solve real problems, integrate models into REST APIs, and handle real-time inference and system automation.
BSc in AI & Cybersecurity
Mansoura, Egypt
Open to Remote
Invisible Technologies Contract
Self-Employed / Freelance Remote
Delta University for Science and Technology — Mansoura, Egypt
Oct 2022 — Jul 2026 | GPA: 3.2 / 4.0
Problem: Organizations need continuous, automated security testing to identify vulnerabilities before real attackers do.
Automated agent that simulates phishing campaigns and cyberattacks for security testing. Runs continuous threat simulations to test system defenses. Helped reduce successful phishing attempts by 25% over three months during evaluation.
Problem: Clinics struggle with patient communication overhead—manual updates, missed appointments, and fragmented records.
Patient communication system using NLP for automated medical messaging. Integrates with the WhatsApp API for real-time updates and two-way engagement. Backend pulls patient data and displays it on physician dashboards showing appointment status and medical history.
Problem: Phishing URLs are increasingly sophisticated, requiring automated detection that goes beyond simple blacklists.
ML system that detects phishing URLs by extracting features from URL structure, domains, and IP metadata. Trained classifiers to flag malicious links with minimal false positives. Deployed as a backend service for automated threat blocking.
Problem: Insider threats and compromised accounts go undetected by rule-based systems that can't adapt to evolving attack patterns.
Multi-model ML system for real-time user behavior analytics and threat detection. Combines anomaly detection, classification, and clustering models to profile normal behavior and flag deviations. Built with a web-based dashboard for security teams to monitor risk scores and investigate alerts.
Problem: Manual code reviews miss critical vulnerabilities, and traditional static analysis tools produce excessive false positives without actionable context.
LLM-powered agent that automatically scans codebases for security vulnerabilities including SQL injection, XSS, and directory traversal. Uses AI reasoning to analyze code context, reduce false positives, and provide developer-friendly remediation suggestions with severity scoring.
Problem: Developing and testing vehicle traction control algorithms requires expensive hardware setups and real-world driving conditions that are hard to replicate consistently.
Physics-based simulation of vehicle traction control systems using Python. Models tire slip ratios, road surface friction, wheel speed dynamics, and torque distribution. Enables rapid prototyping and testing of control algorithms for automotive safety systems in a virtual environment.
Problem: Radiologists face high workloads reviewing chest X-rays, and early signs of cardiomegaly (enlarged heart) can be missed in time-pressured clinical settings.
Deep learning system that analyzes chest X-ray images to detect signs of cardiomegaly. Uses convolutional neural networks to compute cardiothoracic ratios and classify abnormalities. Includes a web interface for clinicians to upload images and view diagnostic results with confidence scores and heatmap visualizations.
Problem: Enterprise systems accumulate thousands of potential vulnerabilities, and security teams lack automated tools to prioritize which threats pose the highest real-world risk.
Machine learning pipeline that classifies software vulnerabilities and scores them using risk assessment models. Extracts features from CVE data, code patterns, and deployment context to predict exploitability and impact. Outputs actionable risk reports with prioritized remediation paths for security operations teams.
I'm open to freelance work, contract roles, and remote positions. If you need an AI engineer who can take a project from requirements to production, let's talk.