Soukaïna Abelhad

Soukaïna Abelhad works at the intersection of AI engineering, technical strategy, and business value. With over eight years of experience across education technology, private equity, telecom, and banking, she has led AI evaluation initiatives, shaped technical architecture, guided build-versus-buy decisions, and delivered AI-powered systems for real-world business workflows.

Her work spans agentic AI, production machine learning, retrieval-augmented generation (RAG), and enterprise automation. She focuses on helping organizations move beyond AI experimentation by identifying valuable use cases, establishing reliable evaluation frameworks, and measuring whether AI systems deliver meaningful operational improvements.

Through her AI automation due diligence framework, she examines what automation metrics actually mean, accounting for human review, autonomous task completion, output quality, and the cost of delivering accepted outcomes.

Building AI systems that can be evaluated

Soukaïna currently leads agentic AI and LLM evaluation at an education technology company. She defined the company's AI evaluation strategy and architecture, aligned product, content, and statistics teams, and drove build-versus-buy decisions for supporting tools. She designed an agentic workflow that generates, evaluates, and corrects assessment content, reducing production cycles from weeks of manual authoring to same-day generation. Her work emphasizes evaluation against domain-expert judgment, reproducibility, and the traceability and auditability of AI-generated content.

Previously, as a Vice President in private equity, Soukaïna developed AI-powered tools for investment professionals, including report-generation assistants, retrieval-augmented generation (RAG) systems over expert-call transcripts, and advisor-matching services. Earlier in her career, she developed fraud detection models in telecom and led cross-functional data science and engineering teams delivering recommendation and churn models in banking.

Across these roles, she combines hands-on AI engineering with technical leadership and strategic decision-making, helping organizations design reliable AI systems, evaluate technology choices, and measure the operational impact of automation.

Making knowledge machine-readable and accessible

Soukaïna has also contributed to open education and knowledge infrastructure. As Data Technical Lead for Wikidata for Education, she contributed to data architecture designed to make curriculum standards and competencies machine-readable, helping connect structured educational information with relevant learning resources.

What an automation percentage leaves out

In her AI automation due diligence framework, Soukaïna examines what claims such as "70% automated" actually measure. A system can generate more output without eliminating the human work required to review, correct, or complete it. Her approach connects three checks:

• Human labor behind automation: Account for the effort required for review, training, escalations, and rework as output volume grows.
• Autonomous task completion: Define clear workflow boundaries and measure how often tasks finish without human intervention.
• Evaluation against expert judgment: Establish whether outputs meet domain-expert standards, rather than relying solely on automated rubrics or model self-evaluation.

Together, these checks establish a more meaningful measure: cost per accepted autonomous task. This accounts for compute, human review, failed attempts, and rework relative to tasks that are completed and meet an explicit acceptance standard.

To make the framework actionable, Soukaïna developed a practical AI diligence card for leaders and buyers evaluating automation claims. The card translates the framework into concrete questions for assessing autonomous completion, human labor requirements, output quality, and operating costs. Access the printable AI diligence card.