Frequently Asked Questions
Only for regulated (GxP) use?
No. For R&D, we right‑size controls—but we design measurements now so you can scale later without rework.
We already have data scientists—why you?
I provide the credibility plan, eval design, and lifecycle controls so leadership can green‑light and QA can defend it.
Vendors say they’re “validated.”
Good—that’s a start. You still own context‑of‑use evidence in your environment. I integrate supplier artifacts into your audit pack.
Who owns validation internally?
QA/CSV + the Business Process Owner. I work in your tenant, align to your QMS, and provide the plan/report and artifacts. Regulatory/Medical advise on claims; IT/Sec covers tenancy and access.
What quality metrics will we get?
Every engagement reports IRA vs. golden set, safety hard-fail rate (e.g., invented citations, prescriptive advice), and rework % (pre/post calibration)—plus drift/monitoring KQIs after go-live.
What’s in the Evidence Pack, exactly?
Intended use & URS • risk register with acceptance criteria • sampling plan & golden set • IRA targets & calibration logs • executed IOQ/PQ with deviations/CAPA • URS↔tests trace matrix • change control (versioned prompts + release notes & rollback) • monitoring log with drift triggers and re-validation outcomes.
Can’t we just pair an ML engineer with a validation engineer?
You’ll get a demo; you may not get an audit-ready evaluation system (risk-tied criteria, calibrated golden set with IRA targets, versioned prompts/RAG, release notes & rollback, and a periodic review plan). That integrated system is what I deliver.
Do you qualify our tools?
Yes. I offer Tool Qualification (TQ-Lite): risk-based Part 11/CSA checks (audit trail/e-sig, traceability, roles/segregation, backup/restore, integration identity) executed in your tenant.
Data & privacy—do you need PHI?
Prefer synthetic/de-identified data. No PHI unless explicitly scoped via DPA. Clinical-adjacent work is non-diagnostic by default unless we agree otherwise.
Do i need to wait until Q1 2026?
No! Discovery and pilots can start now.
Prefer email? Send your use case to kayla@kaylabritt.com (3–5 sentences)