One question across every record, answered with its sources

Transcript

What the walkthrough shows, step by step

The recording has no narration: it shows the product, with captions. Here is what's on screen, with the time each step starts.

1. 0:00

   Projects

   Query Engine projects: a stability program review, a tech transfer to Site B, an annual product review and IND Module 3 pre-submission research.

2. 0:16

   Records in scope

   1,284 records in scope. Ask about anything in your document library; every answer is cited to its source. A quality event: lots SL102-022, SL102-023 and SL102-024 reported CAR-positive cells below the 60% release limit, at 54%, 51% and 56%.

3. 0:22

   One question

   Ask once across every record in scope: batch records, certificates of analysis, reports and deviations. “Three of the last twelve SL102 lots missed the 60% CAR-positive release limit. What is driving it?”

4. 0:28

   Filter

   Narrow thousands of records to the few hundred that could hold the answer.

5. 0:50

   A cited answer

   No single cause is confirmed yet. Four candidates are consistent with the records, each cited: a change of vector lot partway through the series, lower transduction efficiency at Day 1, an incubator temperature excursion logged in DEV-2318, and donor-to-donor variability in the starting apheresis material.

6. 0:56

   A follow-up

   “Compare the vector lots used across those twelve lots. Does anything track with the CAR-positive results?” Sanai reads six vector certificates of analysis, traces the lot genealogy and matches vector lots to release results.

7. 1:02

   Narrowed down

   The follow-up isolates one upstream material shared by every failing lot. The three failing lots are the only ones made with vector lot LV1234-L03, at 0.7 to 0.9 copies per genome against 1.6 to 2.4 for the nine lots made with LV1234-L02. The DEV-2318 excursion affected a lot that passed, so it does not explain the failures.

8. 1:28

   Tone and format

   Set who the answer is for, how deep it goes, and ask for a graph: answers tailored to a research scientist, clinical operations or regulatory affairs.

9. 1:32

   Every lot in one table

   Vector lot, vector copy number and CAR-positive result for each of the last twelve lots: nine conform, and the three made with LV1234-L03 do not.

10. 1:38

    Where in the process

    “Show me where in the process this is happening, and chart CAR-positive by lot.” The process view, from Day 0 to release, pins the loss to a single step: copy number is already low at the Day 3 in-process test.

More videos

Watch the rest of Sanai

• 

  The Sanai film

  1:20

  How Sanai turns raw lab, clean room and clinic data into cited, review-ready regulatory documents: from scattered versions and email threads to a first draft in days.

  Details and transcript→
• 

  Document Authoring

  1:57

  A regulatory document generated in Sanai from a template and the source data, with each claim cited to its source.

  Details and transcript→
• 

  Review Tool

  0:46

  A drafted document checked claim by claim against its sources in the Sanai Review Tool.

  Details and transcript→
• 

  Document Control

  0:47

  Document versions, changes and their traceability tracked in Sanai Document Control.

  Details and transcript→