ML Engineering
Preprocessing, validation, model comparison, failure analysis, and model persistence for NLP, tabular, and security classification tasks.
Hello, I'm
AI Engineer — Security & Governance-Focused
Building AI systems with security, auditability, and deployment discipline.
Get To Know More
My path into AI did not start with AI — it started with circuits. Electronics Engineering gave me a systems-level instinct for how technology works under the hood, which became the right foundation for building and governing AI systems.
Today I work close to critical systems, infrastructure risk, and operational accountability. That context shaped my focus: AI security engineering, retrieval systems, model-serving APIs, evaluation workflows, and governance evidence that can survive real review.
Prompt guardrails, access-controlled retrieval, and safety evaluation.
Model training, validation, API serving, testing, and deployment structure.
Model cards, risk registers, failure analysis, data lineage, and audit notes.
Core Technical
I focus on the parts of AI engineering that matter in real deployment: reliable models, safe serving, measurable retrieval, and documentation that supports governance and audit.
Preprocessing, validation, model comparison, failure analysis, and model persistence for NLP, tabular, and security classification tasks.
Chunking, embeddings, vector search, hybrid retrieval, reranking, source traceability, and retrieval evaluation workflows.
Prompt safety, access control, model cards, risk registers, data lineage, failure analysis, and audit-ready documentation.
FastAPI, Docker, health checks, automated tests, structured logging, CI/CD-ready layouts, and deployment discipline.
Synthetic case study
A hypothetical vendor claims 95% person detection without providing evidence for the intended operating conditions.
Evidence gap. The claim provides no definition of a detection, sample counts, test conditions, or software version.
Proposed verification. Request the test report and agree on the acceptance criteria before a witnessed demonstration. Record missed events, false alarms, and alert delay. Check how operators verify alerts and learn when the feature is unavailable.
Recommendation. Hold technical acceptance of the feature until the evidence and test results support its intended use.
Browse Through My
A focused set of production-shaped AI projects covering AI security, retrieval systems, model serving, evaluation, and governance evidence.
AI Security / Prompt Safety
ShippedA machine-learning guardrail service that classifies prompts before they reach an LLM system.
Model persistence, API serving, health endpoint, automated tests, structured project layout, model card, failure analysis, risk register.
AI Security / ML Engineering
ShippedA binary attack-detection classifier built on network-flow data with validation, feature analysis, and serving readiness.
Chronological validation, attack-family miss analysis, model artifacts, evaluation metrics, model card, limitations notes.
Governance RAG / LLM Systems
ShippedA local retrieval-augmented generation system over the NIST AI Risk Management Framework with source traceability.
Document chunking, top-k retrieval, source citations, traceability requirement, limitation notes.
Secure RAG / Authorization
In ProgressA retrieval system where documents and retrieved context are governed by user roles and authorization boundaries.
Role-aware retrieval, controlled context access, authorization matrix, access-control evidence.
Production RAG / Evaluation
In ProgressA more advanced RAG system moving beyond naive retrieval into hybrid search, reranking, and evaluation.
Retrieval evaluation, ranking comparison, failure inspection, retrieval-quality notes.
MLOps / API Deployment
ShippedA sentiment-analysis model refactored from notebook form into a tested FastAPI service packaged with Docker.
Liveness checks, readiness checks, structured project layout, automated tests, one-page model card.
LLM Infrastructure Study
ShippedA practical benchmark comparing local and hosted LLM usage across latency, throughput, prompt length, and deployment trade-offs.
Transformer Internals
Learning LabAn educational PyTorch implementation used to understand GPT-style architecture blocks, embeddings, and attention mechanics.
Explore My
Get In Touch
Mosab Ahmed
Doha, Qatar