Hadi Hijazi

– Built Fikrah, an AI financial workforce that learns company workflows via screen recordings and deploys supervised agents for finance operations, compliant with Saudi/GCC ZATCA requirements. – Built OsmoseIQ, an AI operations layer for hotels that integrates with existing PMS systems, unifies guest and revenue data, and delivers daily, role-specific operational recommendations. – Architected City, a multi-agent orchestration layer using LangGraph, FastAPI, and Supabase, featuring agent roles, workflow execution, supervision logs, and observability via Langfuse. – Delivered 5 paid client projects, including F&B business intelligence systems and sales automation agents deployed in real operations. – Designed evaluation and supervision layers to ensure correctness, traceability, and controlled autonomy across agent workflows.

– Built an MBSE copilot using Gemini Flash and Neo4j to automate requirement extraction and traceability in aerospace systems, reducing review cycles by ~40%. – Developed a supervised S/4HANA field-mapping agent combining deterministic rules with LLM reasoning, cutting reconciliation workload by ~70%. – Designed self-improving agent evaluation loops using DSPy and Arize Phoenix to score, monitor, and refine model outputs automatically. – Built a video-to-process-flow tool using vision-language models to extract operational workflows from screen recordings.

– Enhanced Bayesian marketing mix models by refactoring priors and feature interactions, improving ROI forecasting accuracy by ~15%. – Built unsupervised behavioral clustering models (KMeans + sequence features) for a multinational automotive client. – Optimized Databricks ETL pipelines, reducing compute time by ~40%.

– Automated multi-entity reporting workflows in Power BI (-50% reporting cycle time).