Wt 5 · Regulatory
Amgen: Evidence Narrative Consistency Under GxP
Triggers: "Walk us through a time you kept an evidence narrative consistent across multiple regulatory documents" (Q16) · "experience with ICH, FDA, regulatory writing standards" (Q23)
S: Amgen needed a GenAI system for FDA regulatory submissions, but any AI-drafted content had to preserve human signatory accountability — no prior GxP-validated precedent existed for this.
T: Lead a cross-functional team (data scientists, regulatory affairs, CMC clinical writers) to architect the first GxP-validated GenAI system FDA would accept for NDA submissions.
A: Designed a multi-agent architecture with a GenAI-to-SCA feedback loop integrating the Veeva Vault regulatory corpus under GxP change control, so every submission improved the template library instead of just consuming it once. This required understanding the scientific and medical writing itself deeply enough to validate that AI-generated output actually matched what the clinical writers needed, not just architecting the system from outside it.
R:
40% reduction in CMC drafting time; scope expanded from Module 3 (CMC) to Modules 4–5 (Safety and Efficacy) after a production milestone triggered Board approval.
Wt 5 · Regulatory
Florida Nuclear Pharmacist Licensure: Authorship With Consequence
Triggers: "medical writer title" gap question (Q24) · general regulatory-authorship credibility
S: Florida's nuclear pharmacist licensure practice standards were inconsistent and needed a defensible regulatory baseline.
T: Co-author a revision to the state licensure regulations governing practice in the field.
A: Drafted the regulatory language directly, working it through public comment before finalization — not reviewing someone else's draft, writing the standard itself.
R:
Became the national model; established licensure as a requirement to practice in the field. Independently checkable public record, not self-reported.
Wt 4 · Business Acumen
Cardinal Health: $300M Recovery Through Data Quality
Triggers: "business acumen" / P&L / executive-scope questions · "translate capability into a recommendation a non-technical stakeholder can act on" (Q22)
S: As Executive Director of Innovation and BI overseeing 80+ radiopharmacies, contract violation claims were going unrecovered due to poor data quality and visualization.
T: Build the analytics case to recover contract violations and establish enterprise master data management.
A: Applied advanced data visualization and Business Objects integration to surface the violations with defensible evidence; separately led an MDM initiative establishing data governance and stewardship.
R:
$300M+ won in contract violation claims; $2M+ in annual cost savings from the MDM initiative.
Wt 4 · Customer Centricity
NewsRx: Catching What Automated Scoring Missed
Triggers: "Tell us about a time your AI system caught a content quality issue that automated checks missed" (Q30) · "where have you seen AI drafting tools fail" (Q19)
S: NewsRx's agentic writing pipeline generates biomedical summaries at scale; automated hallucination scoring alone wasn't catching every real defect.
T: Build a quality framework that catches failures single-metric automated scoring misses, before content reaches a human reviewer.
A: Architected a four-layer QA pipeline: four clinically-specialized LLM judges score summaries blind, cross-correlated against HHEM and BERTScore. Caught a specific case: a summary calling pregnancy outcomes "safe" when one of three women in the study had died — missed by automated scoring alone, caught by the blinded judge layer.
R:
88% reviewer preference for AI-generated content over originals; the specific catch is a concrete, citable failure-mode story, not an abstract claim.
Wt 2 · Medical Writing
Pharmacy School: IND for Radiolabeled WBC Human Injection
Triggers: "medical writer title" gap question (Q24) · direct proof of hands-on regulatory writing, not just governance
S: A pharmacy school project required preparing a formal FDA submission to authorize human use of an investigational radiopharmaceutical: radiolabeled white blood cells for human injection, a real nuclear medicine technique for infection/inflammation imaging.
T: Contribute to preparing the IND (Investigational New Drug application) itself, the actual regulatory document required before human use.
A: Worked through the formal IND content and structure required for FDA authorization of human injection of an investigational biological product.
R:
Direct, hands-on authorship contribution to a real FDA regulatory submission, decades before any of the AI-era work — the honest answer to "have you personally written regulatory content" is yes, this is where it started.
Wt 4 · Technically Adept
J&J MedTech: $750K Migration Under Live Regulatory Load
Triggers: "technically adept" / cloud-scale delivery · "diagnosed a client's failure mode before proposing a solution" (Q21)
S: J&J MedTech needed dual FDA + EU MDR compliance while migrating off Cloudera mid-project, with zero tolerance for interrupted regulatory reporting.
T: Lead the platform migration to AWS while maintaining GxP compliance and uninterrupted delivery over a 2-year window.
A: Built division-specific lakehouses on AWS (S3, Redshift, Databricks), architected integration of SAP and Veeva Vault regulatory feeds into a unified reporting platform for automated EUDAMED submissions.
R:
$750K annual cost reduction, zero interruption to regulatory reporting across the migration window.
Wt 3 · GenAI, keep tight
Optinosis: 3-Person Team Replacing 23
Triggers: "configuring or governing an AI content platform" (Q18) · entrepreneurial/lean questions
S: A prior system required a 23-person team to operate.
T: Rebuild the same operational capability with a lean team using multimodal ML architecture (EHR, imaging, lab data).
A: Founded and architected the replacement system with AI-driven automation replacing manual headcount-heavy processes.
R:
Same capability, 3-person team vs. the predecessor's 23 — keep this one short, it's supporting evidence, not a lead story.
Current pain points to cite – dated, not theoretical
FDA's own AI tool, "Elsa," is reportedly sidelined by FDA reviewers for hallucinating citations and studies, and is mid-forced-migration from Claude to Gemini following a Feb 2026 policy dispute (reported Mar 2026). The regulator writing the AI guidance is currently having the exact problem this role exists to prevent — ask: "given FDA's own tool is reportedly struggling with hallucination and mid-migration between model vendors, how does Cortex/NEXT handle model-version pinning and closed-corpus grounding so a client isn't exposed to the same failure?"
The Lancet, May 2026: 1 in 277 PubMed papers published in early 2026 contain fabricated references, up from 1 in 458 in 2025 (12x in two years) — review articles 57% more likely, and review articles are exactly what feed clinical guidelines and submission narratives. Argues citation-provenance has to be a concrete, auditable mechanism, not an abstract governance promise.
FDA warning letter, Apr 2026 (Purolea Cosmetics): first-ever FDA warning letter naming AI misuse as a cGMP violation — AI-generated specifications distributed with no Quality Unit review of the output. The "no human reviewed it before it left the building" failure mode, now with regulatory precedent.