Belal Zaky

London, United Kingdom

Dubai, United Arab Emirates

Python · propensity-score matching · self-controlled case series · bootstrap CI

• Emulated a new-user, active-comparator study and validated the estimators against a sealed true effect: the crude comparison understated the harm by 46% (OR 1.19), while matching (2.05) and a self-controlled series (2.14) both recovered the truth (2.20).
• A simulation study by design — the only setting where you can score a causal method against a known answer. Same protocol queued for real clinical data (MIMIC-IV, credentialing in progress).

Python · scikit-learn · scispaCy · SNOMED CT · pytest · GitHub Actions · Docker

• Three-stage pipeline: triage a medical abstract for a possible adverse drug event (F1 0.38 → 0.79), extract the drug and effect, then code the effect to SNOMED CT against a 461,000-term vocabulary.
• Built as a system, not a notebook — 32 unit tests, CI on every push, a Dockerfile and a CLI. Chose precision over coverage in coding: a wrong code corrupts a safety signal, a blank one does not.

Python · pandas · SQL · openFDA API · PRR / ROR / chi-squared / Evans criteria

• Built a disproportionality signal-detection pipeline over 20M+ adverse-event reports; validated on the known warfarin–haemorrhage signal (PRR ≈ 5), with negative controls to screen out confounders.
• Across companion projects (a live dashboard and a SQL reproduction), showed how notoriety and sampling distort conclusions — a heartburn drug’s 2020–21 report spike was litigation, not clinical harm.

R (survival) · Kaplan–Meier · Cox PH · Python · scikit-learn · PubMedBERT

• Benchmarked four models for detecting adverse drug events in free text: the domain-tuned transformer won, a keyword model beat a general-purpose one, and a zero-shot LLM came last — domain fit beats model size.
• Modelled time-to-hypertension in 1,100+ patients: diabetics progressed ~2.4× faster, and a seemingly null age effect proved time-varying.

JavaScript · Electron · Node · Markdown · AI agents

• A desktop app that runs my projects, study and deadlines from plain-text notes, with an AI assistant whose drafts I approve; 600+ automated tests, each written after a real failure.
• Notch, a native macOS (SwiftUI) companion app, puts it in the MacBook notch; I use it every day.

Front of house at the New Malden ground: opening up, pitch inspections, directing teams, coaches and referees, and the on-site contact with AFC Wimbledon for delays, cancellations and handovers.

Supported customer service and sales operations in a busy retail pharmacy environment, Dubai.

Coordinated client–team communication and documentation, built digital filing systems, and supported business development and social media.

Ran order fulfilment, customer communication and social-media campaigns, supporting product launches with the founders.

Rep in Year 1 (BSc) and again in Year 2 (MSci). Represent coursemates on the Staff–Student Liaison Committee; consolidated scattered module feedback into one case that reached the department and changed the module.

Represent Pharmacology to prospective students and families at open days, applicant days, campus tours and school outreach.

Selected for King’s flagship leadership programme — employer-led challenges.

Regular prepared and impromptu speaking; earned “Speech of the Meeting”.

Awarded Delegate of the Conference — research, debate and negotiation.