Physician engineer building clinical AI systems that make care safer and more measurable.

Who am I?

I'm Dr. Abdulla Hourani--an MD, machine learning engineer, and researcher working at the intersection of clinical decision-making and rigorous statistical learning. Medicine trained me to reason under pressure and to think in systems: physiology across specialties, how information unfolds over time, and how individual choices propagate into outcomes. Machine learning sharpened that systems mindset into formal tools--probability, statistics, optimization, and generalization. My core interest is building predictive and inferential models that remain trustworthy outside the lab: well-calibrated uncertainty, robustness to distribution shift, principled handling of missingness and irregular sampling, and careful avoidance of leakage--always with an eye on how evaluation metrics translate into real clinical costs.

Much of my work is end-to-end. I build across the full lifecycle--from raw, imperfect EHR data to deployable, monitored inference. This includes data engineering (cohort definition, time-window construction, feature design for longitudinal signals, robust preprocessing for messy hospital data), modeling (classical ML, deep learning, multimodal and sequence-based systems, uncertainty estimation and calibration), and evaluation (proper validation design, sensitivity analyses, ablations, and drift monitoring). I prioritize reproducibility and auditability: pipelines that are versioned, inspectable, and stress-tested. I also care deeply about the "last mile"--turning models into usable tools with safe interfaces, clear outputs, and behavior that remains reliable when deployed in real workflows.

Alongside clinical ML, I work on agentic AI and modern LLM engineering: building systems that can plan, use tools, and execute multi-step workflows--not just generate text. My focus spans training and post-training, including supervised fine-tuning, preference optimization, and reinforcement learning approaches (e.g., RLHF-style pipelines and reward modeling) to shape behavior toward reliability, constraint-following, and task success. I'm especially interested in evaluation beyond surface accuracy: how agents reason under uncertainty, how retrieval and memory should be structured, how to detect and reduce failure modes (hallucination, tool misuse, reward hacking, brittle prompt dependence), and how to measure competence with scalable benchmarks, adversarial testing, and real-world telemetry. Across both medicine and AI, the goal is the same: systems that are not only capable, but measurable--whose strengths, limits, and risks are understood because they are engineered with discipline.

This website is my professional home base: a curated space to share research, publish technical writing, and document lessons across medicine, machine learning, probability, and agentic systems. The blog is where I write practical, implementation-focused analysis on modeling, NLP, evaluation, and the realities of clinical data. The research section organizes projects, publications, and tools--both within Quintessence Health and through broader collaborations. My aim is straightforward: to make my work clear, serious, and easy to engage with--whether you're a clinician, an ML researcher, an engineer, or a founder. If you're building ambitious systems and you care about rigor from data to deployment, I hope this site becomes a useful reference--and an open invitation to collaborate.