Open vacancy · Kestrel Utilities is hiring this

Quality Lead

GMP quality assurance

Here's what AI can do for this role — and what still needs a human. Built straight from Kestrel Utilities’ own job advert, running live on my_db.kestrel_demo.deviations. Not a slide about AI. The job, getting done.

See it run live →Open the live workspace Open the live IGNIS dashboard →

What the AI does

Every line on the left is lifted from Kestrel Utilities’ actual job ad. If a card lacks a harvested JD line, it is omitted. On the right is the AI doing it — with eligible cards running live against the warehouse and offline inspection clearly labelled in the workspace.

What stays human

The honest other half. AI does the analysis; a person owns the decision — especially where regulation, fairness and accountability bite.

Human owns deviation classification and the root-cause conclusion — these carry regulatory accountability. The AI ages and ranks; the quality professional decides severity and closure.
Human owns the effectiveness conclusion and any decision to re-open or escalate. The AI flags the candidates.
Human owns the microbiological investigation and product-impact assessment. The AI screens against limits and prioritises.
Human owns the data-integrity investigation and any GMP documentation remediation — directly inspectable, regulatory-accountable work.
Human owns the disposition decision for every batch (release/reject/hold) — the AI reports the rates, it does not dispose product.

How it works

Ask in English

A plain-English question — the same one the job ad describes — is translated to SQL by the agentic backend.

LIVE — computed now against 27.6M rows

Curated cards run server-side against MotherDuck when eligible. The workspace separately labels any local inspection path.

Real data, live

Runs against my_db.kestrel_demo.deviations. No synthetic numbers.

Self-falsifying

Each figure carries a falsifier — recomputed from the result set, not a stored number, so it can't quietly drift.

Where it plugs in

Function / Meridian surface: Deviations · CAPA · Environmental monitoring · Data integrity · Board pack. Grounded in the real Kestrel Utilities: Meridian — a real-time, event-driven meter-to-cash SaaS platform for energy suppliers · 7M+ accounts · regulated by Ofgem.

Watch it do the job — for real

It's the role getting done: curated questions run live server-side against the warehouse; local inspection is labelled inside the workspace.

Open the live workspace →

Provenance. Synthetic GMP quality dataset (60 batches · 95 deviations · 71 CAPAs · 180 environmental-monitoring samples · 220 batch-record entries). Reproducible — generated by gen-quality-data.mjs (seed 20260608). No real Thermo Fisher or client data. 'Today' is day 160 (2026-06-10); aging, overdue and excursion logic re-derive against it inside each query. Live server-side path: my_db.kestrel_demo.deviations REQUIRES TABLE CREAT

itsorted — I took Kestrel Utilities’ job ads and didn't write a report on what AI could do. I built it. Get the rest sorted →