Mohamed
Ghazal
AI/ML Engineer · Computer Vision · LLM Systems
I build the pipeline between raw data and shipped product — training computer-vision and language models, then wiring them into multi-agent systems and full-stack apps that businesses actually run on.
Data-driven Machine Learning Engineer with a track record of designing, training and deploying production-grade AI systems — from computer-vision architectures to LLM pipelines. I optimize model architectures, build robust engineering pipelines, and develop enterprise-scale multi-agent frameworks, drawing on a Software Engineering background to bridge ML research and scalable application deployment.
Three roles, one thread: turning models into working systems.
- Architected data-analysis modules for a Governance, Risk & Compliance (GRC) enterprise system, cutting manual threat-assessment work.
- Optimized application runtime performance by 15% through behavioral-pattern analysis and high-throughput data processing.
- Built responsive dashboards (HTML5, CSS3, JavaScript) surfacing complex algorithmic metrics and predictive logs.
- Engineered automated Python (Pandas, NumPy) pipelines to clean and preprocess 100K+ unstructured records.
- Ran deep exploratory data analysis and built predictive models that improved trend-forecasting accuracy.
- Synthesized mathematical models into business-ready documentation to drive data-informed decisions.
- Developed a deepfake-detection system with custom CNN architectures for facial-manipulation forensics.
- Built a low-latency MTCNN preprocessing pipeline for real-time face extraction across video streams.
- Used OpenCV for frame-level processing, speeding up feature extraction under compute constraints.
Selected builds, filtered by what's under the hood.
AI Compliance & Fraud Co-Pilot for SMEs
June 2026 – PresentBuilt a multi-agent compliance system giving SMEs an affordable, explainable alternative to enterprise GRC tools for verifying ID documents against Turkey’s KVKK data protection law.
Engineered a CV forensics module combining Error Level Analysis, MTCNN face validation, and content moderation to automatically flag tampered or invalid document submissions.
Architected a 4-agent LangGraph pipeline grounding LLM-generated audit reports in a RAG-based legal knowledge base, and validated it against a synthetic test suite for accuracy.
Enterprise Multi-Agent Orchestration Framework
Apr–May 2024A production-ready multi-agent pipeline that automates complex, multi-step corporate operations through autonomous, asynchronous agents.
Designed a centralized state-management layer with a supervisor-agent architecture, improving error recovery and task routing, and wired in secure API gateways for safe transactional queries against SQL Server.
ASL Recognition System
Apr–Jun 2025Graduation project — a real-time American Sign Language recognizer built on MediaPipe hand tracking and spatial coordinate modeling.
Trained a Deep Neural Network with batch normalization in TensorFlow/Keras and optimized it with TensorFlow Lite for under-50ms web inference, served through a Flask backend with an async JS frontend.
AI Medical Assistant
Jan–Mar 2026An intelligent health-guidance platform: a context-aware medical QA engine using Retrieval-Augmented Generation over dense medical reference literature.
Added a ChromaDB vector-search layer to cut query response time and reduce hallucination, plus a Flask dashboard for dynamic PDF ingestion and real-time summary generation.
Deepfake Detection System
Apr–Aug 2023Built at INOSAS: a custom CNN architecture for classifying facial-manipulation forensics with an MTCNN-based real-time face extraction pipeline.
Used OpenCV for frame-by-frame processing to speed up feature extraction under compute-constrained conditions.