Can Katalyze digitize our paper batch records?

Yes. Paper batch records are where teams typically start with Document Intelligence. It reads the printed form, the handwritten values, the corrections and the margin notes on every page and turns them into structured data in your schema, each value linked to the page it was read from, with anything it is unsure of flagged for your experts before the data moves on.

What documents can Katalyze process?

When Document Intelligence is the primary purchase, teams commonly begin with paper batch records, certificates of analysis, PDFs, scans and records received from suppliers, CROs, CDMOs and labs. Within an Agentic Solution, the document set is selected for the operational job and may include recipes, process records, deviation records, SOPs and transfer documentation.

How does Katalyze handle tables and handwriting?

Katalyze can extract information from text, tables, symbols, handwritten fields and margin notes. Performance depends on the document type and image quality, so extracted information can be routed through defined checks and expert review before downstream use.

Can reviewers verify extracted information?

Yes. Extracted information remains linked to its source so reviewers can inspect the underlying evidence and resolve flagged exceptions.

Does Document Intelligence replace our existing systems?

No. Customer systems remain the source of truth. Katalyze can deliver structured outputs to configured destinations or combine document information with relevant system data and external scientific sources through the Context Layer.

Is Document Intelligence a standalone product or part of a larger solution?

It can serve either role. Teams can begin with Document Intelligence to structure and review high-value records such as batch records and certificates of analysis. It also provides document-derived context within Agentic Solutions such as Deviation Investigation, Yield Optimization, Raw Material Characterization and Tech Transfer.

What is the difference between Document Intelligence and the Context Layer?

Document Intelligence extracts and contextualizes information from operational documents. The Context Layer combines that information with data brought in through Connectors and organizes it using a customer-specific Ontology and a Knowledge Graph of relationships.

Document Intelligence for Pharma

Stop searching through batch records and certificates of analysis. Start asking them questions.

Document Intelligence by Katalyze turns batch records, certificates of analysis, recipes and process records, lab reports, deviation records, SOPs and transfer documentation, on paper, in scans and in PDFs from your labs, suppliers, CROs and CDMOs, into structured, source-linked data your teams can review, analyze and trust.

Book a Demo See How It Works

The records

Batch record · Certificate of analysis · Master batch record

In your schema

Feed lot2231-B, correctedRead from feeding record
Feed 3, recorded11:40Read from feeding record
Margin notefeed 3 added late, see DV-0088Read from page margin · Review
Certificate0091Read from certificate header
Feed 3, planned09:00Read from master batch record

Where it goes

Context Layer
Deviation Investigation · DV-0088
Yield Optimization
Your systems

Today

The records are there. Using the data still takes work.

Batch records, certificates of analysis, recipes, lab reports and deviation records hold the information your teams need, but understanding a complete record and its meaning still requires experts to find the relevant values, interpret the tables and notes, and transfer the information into another file or system.

Binder · Site A · P-14p. 6 of 12

Inbox2026-02-27

FromQC release, the supplierSubjectCertificate of analysis, lot 2231-B

Please find attached the certificate for feed lot 2231-B, shipped today. Certificate 0088 for lot 2231-A went with the earlier shipment.

Certificate of analysis

certificate-0091.pdf1 page · 212 KB

P-14 batch tracker.xlsxthe team's drive

	A	B	C	D
1	Batch	Started	Feed lot	Harvest
2	BR-0410	2026-02-02	2231-A	2026-02-16
3	BR-0411	2026-02-16	2231-A	2026-03-02
4	BR-0412	2026-03-02	2231-B	2026-03-16
5

Quality systemDeviation DV-0088

Batch: BR-0412 · feed 3
Title: Feed addition later than planned
Description: Feed 3 was added at 11:40 against the planned 09:00; noted in the record's margin.
Raised: AM · 2026-03-07
Status: Open · under investigation

Document systemRev 3 · controlled

P-14 master batch recordSection 6 · Feeding schedule

Step	Feed	Day	Planned
6.1	Feed 1	Day 3	09:00
6.3	Feed 3	Day 5	09:00
6.6	Feed 6	Day 8	09:00

Effective 2026-01-15

Shared drive / QC / BR-0412

Harvest results

SampleBR-0412 · harvestReceived2026-03-16

Test	Method	Results
Harvest titer	HPLC	p. 2
Sampling stage	Harvest	p. 2

AnalystReviewer

Today, your experts

Find the values in a long record
Interpret the tables, symbols and notes
Re-key them into another file or system

With Document Intelligence

Every value is read into your schema, linked to its page
What it is unsure of is flagged for your experts
The data is ready for your systems and Agentic Solutions

Free your experts to advance the science.

See it work

Digitize and Connect Your Data

See a batch record become usable data.

On the left, the record as it arrived: printed form, handwritten values, a correction, a margin note, scribbles and checkmarks. On the right, the same page digitized into your schema: every value with what the form called it, where it was read and what still waits for review.

Ask your data a question

Document Intelligence · BR-0412page 6 of 12 · 18 fields · 1 for review

Field	Read from	Value	State
Header
ProductProduct	p. 6 · header	P-14 mAb
Batch numberBatch No.	p. 6 · header	BR-0412
BioreactorBioreactor	p. 6 · header	2000 L
Date startedDate started	p. 6 · header	2026-03-02
Feeding record, section 6
Feed 1, recorded timeRecorded time	p. 6 · step 6.1 · handwritten	09:10 · planned 09:00
Feed lotFeed lot	p. 6 · step 6.1 · corrected, AM 03-05	2231-B · corrected from 2231-A
Feed 2, recorded timeRecorded time	p. 6 · step 6.2 · handwritten	09:05 · planned 09:00
Feed 3, recorded timeRecorded time	p. 6 · step 6.3 · handwritten	11:40 · planned 09:00
Feeds 4 to 6, recorded timesRecorded time	p. 6 · steps 6.4 to 6.6 · handwritten	09:00 · 08:55 · 09:05 · planned 09:00
Feeds verifiedVerified by (initial / date)	p. 6 · steps 6.1 to 6.6	KS · every feed
Rows 6.7 to 6.8(rows crossed through)	p. 6 · rows struck, KS 2026-03-16	Not used
Harvest sample to QCHarvest sample to QC (date / initials)	p. 6 · harvest line · handwritten	2026-03-16 · AM
Titer resultTiter result: see Harvest results (analytical record)	p. 6 · printed line under the harvest sample	See Harvest results
Review
Deviation issuedDeviations issued against this record	p. 6 · deviations line	DV-0088 · AM 2026-03-07
Margin note(margin)	p. 6 · margin, KS 2026-03-16	feed 3 added late, see DV-0088	Review
Manufacturing reviewManufacturing review complete (sign / date)	p. 6 · review block	Signed · 2026-03-17
QA reviewQA review (initial / date)	p. 6 · review block	KS · 2026-03-18
No further correctionsNo further corrections on this page.	p. 6 · footer, KS 03-18	Yes · box checked

An illustrative batch manufacturing record page, filled in by hand: a feeding record table, a corrected lot number, a margin note and a signature

page 6 of 12 · 18 fields · 1 flagged for review

The questions

Ask your data a question. The answer is sourced and connected thanks to Document Intelligence.

Once the records are data, your teams ask Katalyze questions in plain language. An answer can draw on several records at once, and it cites each one it was read from, so an expert can check it against the page.

Book a Demo

Ask Katalyze · P-14Answered

Question

Which P-14 batches recorded a feed addition later than planned, and how did their harvest titer compare?

Answer

Two of the twelve in the comparison set: BR-0412 and BR-0421. On both, the feeding record shows feed 3 recorded later than the master batch record planned; on BR-0412 at 11:40 against 09:00, with a margin note referring the late addition to DV-0088.01

Both recorded lower harvest titer than the other batches in the set. BR-0415, fed as planned, was comparable.02

All three used feed lot 2231-B, under certificate 0091. The comparison does not separate feed timing from the lot.03

Follow-upWhat does DV-0088 record so far, and is it still open?

The workflow

How it works: contextualized data extraction.

A workflow made from reusable extraction, assembly, rules, review and egress steps, built from scratch for the pharmaceutical manufacturing industry.

Explore the Context Layer

01 · ExtractBatch record BR-0412 · p. 6

02 · AssembleIn your schema

03 · Check5 rules · 1 for review

04 · ReviewFlagged · beside its page

05 · Deliver5 fields · source-linked

010203

01Feed lot2231-B, correctedfeeding record
02Feed 3, recorded11:40feeding record
03Margin notefeed 3 added late, see DV-0088page margin

Batch recordBR-0412 · 12 pages
Certificate of analysislot 2231-B · PDF
Master batch recordP-14 · Rev 3 · PDF

Field	Value	Read from
Feed lot	2231-B, corrected	feeding record
Feed 3, recorded	11:40	feeding record
Margin note	feed 3 added late, see DV-0088	page margin
Certificate	0091	certificate header
Feed 3, planned	09:00	master batch record

Field	Value	Rule	State
Feed lot	2231-B, corrected	Lot on record
Feed 3, recorded	11:40	Deviation if late
Margin note	feed 3 added late, see DV-0088	Refers to a deviation	Review
Certificate	0091	On file for the lot
Feed 3, planned	09:00	Current revision

Margin notefeed 3 added late, see DV-0088page marginReview

KS · 2026-03-18Checked against DV-0088, still openConfirmed

Field	Value	Read from	State
Feed lot	2231-B, corrected	feeding record
Feed 3, recorded	11:40	feeding record
Margin note	feed 3 added late, see DV-0088	page margin	Confirmed
Certificate	0091	certificate header
Feed 3, planned	09:00	master batch record

Context Layerwith the data from your systems
Deviation Investigation · DV-0088
Yield Optimization
Your systems

Document intelligence for pharma

Designed for pharmaceutical records.

Work with long, complex documents containing tables, diagrams, scientific notation, handwriting and inconsistent formats. Batch records and certificates of analysis are where teams commonly begin; within an Agentic Solution the document set is selected for the job.

Book a Demo

Batch record

The executed record of one batch: printed forms filled in by hand across dozens of pages, with corrections struck once and initialed, notes in the margin and a signature on every review.

What is read

Header fields
Process tables
Hold times
Yields
Corrections
Margin notes
Signatures

FeedsDeviation Investigation · Yield Optimization

Explore batch records

Built for GxP environments

Traceable by design.

Document Intelligence by Katalyze works within your organization's validation and approval framework: extracted information stays linked to the original page, flagged exceptions go to your experts before anything moves downstream, and every run is recorded. Nothing is re-keyed by hand, nothing uncertain is filled in, and nothing flagged moves on without a person. Built for GxP environments and processes.

How compliance works

Document run · BR-0412 · 2026-03-1806 Documented processing

Recorded

09:12:0409:12:3109:12:4009:58:1209:58:30

12 pages readBatch record BR-0412, a paper scan: 18 fields, 1 table, 2 corrections and 1 margin note, each keeping the page and the region it was read from12 pages
01 Source-linked dataInspect the original evidence behind extracted information.
Placed, not re-keyedEvery value in the field your operation defined for it, from the page rather than a keyboard; the corrected feed lot and one deviation reference resolvedYour schema
02 Customer-defined schemasInformation organized around the terminology, document structures and field definitions your operation uses.
Unsure, so flagged17 values pass their rules; the margin note, which the read is unsure of, is flagged instead of filled in1 flagged
03 Uncertainty surfacedMissing or uncertain information is surfaced instead of passing silently.
A person decidesKS checks the flagged note against DV-0088 and confirms the read; the decision is recorded with the time and the initials, and the deviation stays openConfirmed
04 Expert controlReview extracted information and flagged exceptions before it moves downstream.
Lineage keptThe reviewed record reaches the Context Layer with every value's source and every decision, inside your validation and approval frameworkDelivered
05 Designed for validated processesGoverned outputs used within your organization's validation and approval framework.

Where the data goes

Connect the records to the rest of your operation.

Agentic Solutions use this source-grounded context to establish ground truth, investigate with evidence and execute governed work. Start with the documents and data needed for the first operational job; the same Context Layer can be reused and expanded as additional solutions are introduced.

All Agentic Solutions

01
Agentic Solution
Deviation Investigation
Assemble and validate relevant evidence to investigate possible causes.
Explore Deviation Investigation
02
Agentic Solution
Yield Optimization
Connect recipes, process records, material data and system data to investigate yield performance.
Explore Yield Optimization
03
Agentic Solution
Raw Material Characterization
Connect material, supplier and scientific information to investigate variability.
- Certificates of analysis
- Lab reports
Explore Raw Material Characterization
04
Agentic Solution
Tech Transfer
Assemble process knowledge and supporting evidence to prepare transfer documentation and action plans.
Explore Tech Transfer

In practice

Scientists and engineers use Katalyze to investigate, improve and advance.

What teams started with, what their records became and what changed for the scientists and engineers who use them.

See customer results

Cell therapy developer

Batch record review went from days to hours.

A cell therapy developer started with paper batch records and certificates of analysis from three suppliers. The records became source-linked data in its own schema, and deviation investigations now start from the data, not from the binder.

See customer results

Every value we used to re-key by hand is read from the page with its source beside it. Our scientists spend the time on the science instead.

Head of MSATCell therapy developer

less time re-keying batch records

to review a batch record, from days

Contract manufacturer

Certificates of analysis from forty suppliers, one schema.

A contract manufacturer receives certificates of analysis in forty layouts. Each is read into the same schema on arrival, checked against the specification and linked to its lot, so material release starts with the data already in place.

See customer results

Forty suppliers, forty formats, and one table our quality team reads the same way every morning.

Director of QualityContract manufacturer

suppliers' certificates read into one schema

material release from receipt, from a week

Biologics site

Transfer documentation assembled in weeks, not quarters.

A biologics site moving a process between sites started with recipes, process records and SOPs from both ends. Every parameter is read with its source, so the transfer package is assembled from the records rather than retyped from them.

See customer results

The transfer package used to be a quarter of retyping. Now it is the records themselves, side by side, with every number pointing back to its page.

Tech Transfer LeadBiologics site

faster transfer package assembly

values retyped by hand

FAQ

Questions about Document Intelligence.

What it reads, how it handles tables and handwriting, and where the data goes. The full set of answers is in the FAQ.

All questions and answers

Book a demo

Put the information in your records to work.

Bring the job you want to move forward and the records you'd like connected. Thirty minutes on how Katalyze would do the work, and where your experts come in.

Or, just a quick chat to run through the product, your call.

Stop searching through batch records and certificates of analysis. Start asking them questions.

The records are there. Using the data still takes work.

See a batch record become usable data.

Ask your data a question. The answer is sourced and connected thanks to Document Intelligence.

How it works: contextualized data extraction.

Designed for pharmaceutical records.

Batch record

Traceable by design.

Connect the records to the rest of your operation.

Deviation Investigation

Yield Optimization

Raw Material Characterization

Tech Transfer

Scientists and engineers use Katalyze to investigate, improve and advance.

Batch record review went from days to hours.

Certificates of analysis from forty suppliers, one schema.

Transfer documentation assembled in weeks, not quarters.

Questions about Document Intelligence.

01Can Katalyze digitize our paper batch records?

02What documents can Katalyze process?

03How does Katalyze handle tables and handwriting?

04Can reviewers verify extracted information?

05Does Document Intelligence replace our existing systems?

06Is Document Intelligence a standalone product or part of a larger solution?

07What is the difference between Document Intelligence and the Context Layer?

Put the information in your records to work.