Ziad Nawar

Datacation is an AI and data science consultancy company.

• •Productionized computer vision models end-to-end — GPU selection, quantization, camera/PLC integration, ByteTrack, and evaluation pipelines — delivering a semi-production deployment for a client.
• •Designed and deployed LLM-powered solutions including RAG systems, agentic workflows, document chat, and intelligent data extraction from Excel/XML using structured output, prompt optimization (DSPy), and LLM-as-a-judge evaluation.
• •Architected and deployed a production-grade document processing pipeline on Azure (Functions, Durable Functions, MLflow monitoring, Databricks), optimizing for accuracy → speed → cost and handling LLM edge cases at scale.
• •Led full-stack web development of the company website using Next.js, Payload CMS, and MongoDB — including multi-language support, Vercel deployment, Azure Blob storage, caching/revalidation strategies, SEO, and Google Analytics integration.
• •Delivered multiple client projects (AWL, GPH, Virtual Vaults, Flanderijn, CRH) covering demand forecasting, NER, product recommendations, and AI-driven document intelligence — managing stakeholder relationships and proposal writing.
• •Spearheaded AI strategy sessions with C-level stakeholders — meeting with CPO/CTO to strategize AI adoption, competitive analysis, and product roadmap across client organizations.
• •Established engineering best practices across the team — code review sessions, Git workflows, pre-commit hooks, SRP/DRY principles, and proper architecture planning before implementation.
• •Designed and completed a comprehensive 6-week onboarding plan for new joiners covering technical, non-technical, and business pillars with absorb, watch, and contribute phases.
• •Mentored juniors and mediors — guiding team members in Azure, Databricks, AI maturity, blog publishing, and first production deployments, while conducting 10+ technical interviews for hiring.
• •Built internal tooling including an LLM coding assistant, automated YouTrack-HubSpot integration, and a file categorization system using document clustering for data room analysis.

Artefact is a consultancy company specialised in AI and digital marketing.

• •Collaborated within a team to analyze marketing performance for a multi-billion-dollar retail client in the US market, leveraging causality analysis and time-series forecasting techniques.
• •Led the end-to-end development and deployment of a Bayesian-based ranking model for a popular e-sports game, including research, planning, implementation, and delivery.
• •Mentored an intern over a 5-week period, providing technical guidance and ensuring timely completion of an internal solution aligned with project goals and deliverables.

Grant Thornton is a global audit, tax, and advisory firm.

• •Collaborated with non-technical colleagues, translating business challenges into actionable technical solutions.
• •Delivered multiple successful full-stack projects using Outsystems, including the Tax Reporting System.
• •Designed and implemented a customized surveying system, enhancing workflows.
• •Developed systems for financial document parsing and invoice generation automation.

ScenWise is a software development company works in the mobility sector.

• •Contributed to a traffic data analysis web application back-end using Java, REST API, Spring framework, and database technologies.

This thesis addresses the issue of reliability in machine learning (ML) systems used for computer vision applications, which can fail when faced with slightly different data than their training set. The author proposes a system to help ML practitioners debug their computer vision models before deployment. The system utilizes human computation operations to identify expected model behaviors, compare them to actual model behaviors, and assess model performance. The author claims that their system is the first to allow ML practitioners to define debugging goals, configure debugging sessions, and automatically generate model debugging reports. They conducted a comprehensive evaluation of the system, demonstrating its correctness, informativeness, and cost-effectiveness, even when considering potential human errors. Despite some limitations, the work represents an important step towards assisting practitioners in debugging ML models, encouraging further optimization, and providing open access to their code for broader usage and experimentation.

The paper discusses the collaboration between humans and digital computers in various tasks, emphasizing the importance of combining artificial intelligence (AI) agents with humans to improve team effectiveness. The focus is on the human-AI agent team and the role of shared mental models in team performance. The paper aims to experimentally analyze how different shared mental models impact human-AI agent collaboration. The hypothesis is that exchanging more information within the shared mental model between humans and AI agents leads to higher team performance. The paper outlines the research questions, subdivides the research problem, and describes the experimental design. It also presents results and statistical analyses, touches on ethical considerations, and provides a discussion of the experiment's outcomes and future research recommendations before concluding.