Batch data analysis in plain language — every answer with its query, its lots or its page
Batch analysis that took three months now takes hours, with the query behind every answer.
See one case, screen by screen ↓A harvest comes in under the release limit, and the questions start. Is it one batch or a trend? Which raw material lots went into it, and which batches still in expansion share them? What does the procedure ask for, and by when? The answers sit in the batch record, the lab results, the material movements and the SOP library — a dataset of more than 300 columns and thousands of lots.
So the analysis is done by hand: exports from the MES and the lab system, a spreadsheet lined up batch by batch, the genealogy walked one lot at a time, the procedure looked up and quoted. It takes months, it is easy to get wrong, and by the time the shared lot is found the next batch has already been fed from it.
Estimated split for a typical batch analysis on one line, by hand and with the solution — hours of work instead of months.
Five specialist agents route, query, trace and cite every plant question; a person confirms before it is saved.
CRV-26-0431, a Corventa batch on Line C2 at Plant 3 — Leiden, was harvested at 68.9 % viability — under the 70 % release limit. Here is what the MSAT process scientist does next, screen by screen, in the working solution.
Lena opens the solution: 7 batches in progress, 6 harvested in the last 30 days with 1 below the release limit, 4 open deviations, and one raw material lot on watch — culture medium MED-24-0917. First under “Needs you”: CRV-26-0431, harvest viability 68.9 %, QA disposition needed, genealogy ready.
Read-only sources: batch data · lot genealogy · procedures library.
She asks: “Show viability at harvest for the last 20 batches and highlight outliers.” The batch data specialist finds viability at harvest — lab test VIAB-HARV, 7-AAD flow — among 312 columns, runs a read-only query and checks the control limits. The answer: an average of 87.6 %, and two batches under the lower control limit of 83.4 % — CRV-26-0431 at 68.9 %, under the 70 % release limit, and CRV-26-0418 at 72.4 %, under the 75 % action limit.
“Outlier rule: below mean − 3σ of the 32 batches before Jul 21 (83.4 %).”
The query sits beside the answer: a SELECT over the MES batch table and the lab results, Corventa, Line C2, cancelled batches left out — 20 rows, 3 of 312 columns, read in 0.8 s. Lena reads it, then confirms the answer. Her name and the time go on the answer and into the audit trail.
Answers show their query or page. A person confirms before an answer is saved or shared.
A click on the 68.9 % point opens the batch genealogy: apheresis CV-4471-LD through selection, transduction and expansion to the drug product, Corventa, 2 bags — with every raw material lot attached. The genealogy specialist walks 14 links, compares the lots with the other outlier and finds one in common: culture medium MED-24-0917. The same lot is in CRV-26-0436, in expansion now, on day 6.
“Same supplier, same medium. Lot MED-24-0917 came in at 318 mOsm/kg osmolality — the top of its 280–320 range.”
Lena asks, on the batch: “What does the procedure say when harvest viability is below 75%?” The procedure specialist searches 642 documents and cites SOP-MFG-0214 v7.0 §8.3 — the effective version; the v8.0 draft is not cited. Tell MSAT and QA within 4 hours, open a deviation the same working day, keep two 1 mL post-harvest samples at ≤ −150 °C. Two more citations complete the answer: below 70 %, no release without a QA disposition decision (SOP-QA-0107); below 82 % in process from day 6, check the raw material lots before the next feed (SOP-MFG-0211).
One click on the citation opens SOP-MFG-0214, Harvest, formulation and cryopreservation, at section 8.3 on page 12, with the passage highlighted. Beside it: effective v7.0, owned by MSAT, approved by QA, cited 37 times in the last 30 days.
“The highlighted passage on page 12 is the one the assistant cited.”
The procedure asks for a deviation today, so Lena asks for a draft. The deviation drafter reads the genealogy and the three batch results and fills the form from SOP-QA-0102: the batches affected, what happened — two of the last 20 harvests below the lower control limit, both on MED-24-0917 — and the immediate actions: quarantine the remaining 14 bottles, feed CRV-26-0436 from lot MED-24-0952 at 14:00 once QA approves. It proposes Major, and says Dana Okafor decides.
“Drafted by an agent from the cited records. Nothing is sent until you sign; QA decides the class and the quarantine.”
Lena signs DEV-2026-0187 with her password. The meaning of the signature is recorded with it: “I raise this deviation and confirm the facts are correct.” The investigation is due Nov 5, 2026 — 30 days.
DEV-2026-0187 goes to Dana Okafor, who classifies it. Omar Haddad, the Qualified Person, decides on the quarantine of MED-24-0917. On the genealogy the lot now reads “Quarantine requested”, and CRV-26-0436 shows “Hold requested” — its next feed to come from lot MED-24-0952 once QA approves.
The dashboard shows batch analysis down from about 90 days by hand to a median of 5.5 hours this quarter, questions answered per week rising from 112 to 154, and 86 % of 660 answers confirmed by reviewers. Corrected answers go back to the specialist as examples. Under “Caught early”: CRV-26-0436, found on day 6.
The same low harvest, seen by the five people who carry it — what their week looked like, and what it looks like now.
Five specialist agents route, query, trace and cite every plant question; a person confirms before it is saved.
Reads each question and routes it to the right specialist — batch data, genealogy or procedures — asking one clarifying question when the time range or batch is unclear.
Turns plain questions into read-only queries over 312 batch columns, uses the approved metric definitions, flags outliers against the control limits, picks the right chart and shows the query.
Traces a lot from raw material to finished product in both directions and finds lots shared between batches — including batches still in process.
Answers process questions from SOPs and batch records, quoting limits exactly as written, with the section and page for every statement.
Drafts a deviation from the genealogy and the cited records, following the deviation management SOP, and proposes a classification and immediate actions.
Confirms before save or share. The agents propose; a named person decides.
MSAT, QA and manufacturing ask about results, lots and procedures on their line. The right specialist answers — with the chart, the query or the page.
Compare selection yield by apheresis center for Q3
Selection yield in Q3 averaged 65 % across 24 batches. Brightwater Cancer Centre is lowest at 54 % (3 batches) — its material arrives after a median 31 h transit, against 18 h for the others. Every center is above the 40 % minimum in the batch record.
Which batches used vector lot LVV-2238?
13 batches, CRV-26-0412 to CRV-26-0424. Vector copy number stayed between 1.8 and 2.9 copies per cell, inside the 5.0 limit on every batch. The lot is fully used; the next lot, LVV-2241, started with CRV-26-0425.
Does seeding density affect fold expansion at day 9?
Yes, up to a point. Fold expansion at day 9 rises from 38× at the lowest seeding band to 52× at 0.9–1.1 × 10⁶ cells/mL, then levels off. The batch record target of 1.0 × 10⁶ cells/mL sits in the best band.
Tell me when harvest viability drops below 82% on any batch
Done — a new watch rule: harvest viability below 82 % notifies you and the line lead within the hour. It reads the lab results every 15 minutes, read-only. You can switch it off in Settings.
13 screens from the working solution, on its sample data. Pick one to see it large.
What each specialist may read, who approves what, and the watch rules that tell the right people when a limit is crossed.
| Setting | Default | Choose from |
|---|---|---|
| Batch data the specialists may read | Line C2 only | Line C2 only · All Corventa lines |
| Lot genealogy the specialists may walk | Plant 3 — Leiden | Plant 3 — Leiden · All plants |
| Cite effective procedure versions only | On | On · Off |
| Show the query with every answer | On | On · Off |
| Mask patient identity | On | On — locked |
| Raw material lot hold approved by | Omar Haddad | Omar Haddad · Dana Okafor |
| Deviation classification decided by | Dana Okafor | Dana Okafor · Omar Haddad |
| Charts saved to dashboards confirmed by | Sam Patel | Sam Patel · Lena Ortiz · Dr. Maya Chen |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample plant data · “proven” = in user acceptance testing on a commercial cell-therapy line at a global medtech/pharma manufacturer · “target” = the design goal, measured in the live solution · “estimated” = our estimate. People, companies and products named on this page are fictional — characters and sample data in the working solution.
A way for MSAT, QA and manufacturing to ask plain-language questions about yield, deviations and batch genealogy, answered from the plant’s own systems. Five specialist agents route each question, query the batch data, trace the lots and cite the procedures; a person confirms an answer before it is saved or shared.
The batch data specialist turns the question into one read-only query over the batch record and lab result columns, using the approved metric definitions — for example, viability at harvest is lab test VIAB-HARV. It flags outliers against the control limits, picks a line chart, bars or a table, and shows the query and the tables it read beside the answer.
Yes. The genealogy specialist walks the lot links in both directions — from the drug product back to every raw material and component lot, or from a lot to every batch that used it. When asked about an outlier, it compares its lots with the other outliers and lists batches still in process that use the same lot.
No. Every connection is read-only, and data queries are SELECT only with a 10,000-row limit. In the quality system the deviation drafter can only prepare a draft; nothing is sent until a person signs it.
Every data answer shows the query it ran, and every procedure answer cites the effective document by section and page. A person confirms the answer before it is saved or shared, and corrected answers go back to the specialist as examples. In user acceptance testing on a commercial cell-therapy line, accuracy was above 80 %.
Yes. The person raising a deviation signs it with password and the meaning of the signature, QA decides the classification, and quarantine of a raw material lot needs the e-signature of the named approver. Who approves each is a setting.
Patient identity is masked: only chain-of-identity codes are shown, never names or dates of birth, and the setting is locked on.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your batch data definitions, lot genealogy, procedures and deviation SOP — and it goes live once every quality gate has passed. We will walk you through it on a sample of your own batches first.
Any batch-record question answered in seconds, cited to the exact page and step.
1-min filmValidation DocumentsValidation protocols and reports drafted in minutes, every value traced to its source page.
1-min filmDeviation & CAPAEvery deviation reaches a ranked, evidence-backed root cause in days, closed inside the 30-day target.
1-min filmWe’ll run Ask the Plant on a sample of your own batch data, lots and procedures.