Evaluation Before Automation
I start with what must be proven: evidence recall, decision correctness, exception behavior, citation faithfulness, audit trail quality, and conditions for abstention.
Profile
Research-trained AI systems work for regulated, bilingual enterprise environments.
My background combines a PhD in Electrical Engineering, first-author research in computer vision and optimization, and delivery experience across insurance, compliance-oriented automation, and bilingual Arabic-English enterprise environments.
Operating range
Where the work is strongest.
Regulated Workflow Judgment
Insurance and compliance work require authority boundaries, human review, policy interpretation, exception handling, and a record of why a decision was made.
Arabic-English Enterprise Reality
Bilingual systems fail in specific ways: translated clauses drift, local terminology matters, dialect changes intent, and document hierarchy can decide the answer.
Research-Trained ML Judgment
Doctoral research and first-author publications shaped a habit of controlled comparison, reproducibility, and skepticism toward unsupported model claims.
2025 - Present
Insurance AI and Digital Transformation
Applied AI work for insurance and regulated workflows in Saudi Arabia, focused on production readiness, governance, human-in-the-loop design, and operational measurement.
2023 - 2025
AI Venture Building and Consulting
Co-founded and advised applied AI initiatives across autonomous systems, computer vision, technical handover, and partner-facing product roadmaps.
2024 - 2025
AI-Enabled Operations
Applied computer vision and automation to site operations, reporting, and safety workflows, with emphasis on operational visibility and repeatable handover.
2017 - 2022
PhD Researcher, Electrical Engineering
Developed optimization and ensemble-learning methods for medical image classification, producing first-author peer-reviewed publications and reproducible experimental pipelines.
Research
Published machine-learning research behind the systems work.
Doctoral work on ensemble optimization and medical image classification established the research habits that still matter in production AI: controlled comparison, reproducibility, and careful claims about model behavior.
Method
Principles I use to judge AI systems.
- Evidence is part of the product, not an afterthought.
- Human review should be designed as a control, not inserted as a vague fallback.
- Arabic fluency is not enough; evaluation must test meaning, evidence, and decision behavior.
- Agentic systems need session integrity: calls, arguments, state changes, permissions, and outputs must be replayable.
- The best AI architecture reduces institutional uncertainty rather than hiding it behind a fluent interface.