Introducing

Kriyam DocForensics. Verify every document. Trust every decision.

An agentic document forensics pipeline for images, scans and PDFs — detecting pixel, font, metadata and arithmetic anomalies, deepfakes and identity fraud.
Every flag marked on the page. Every verdict explained.

Document received agent · docforensics-v2
bank-statement.pdfPDF
Account statementPage 1 of 3
02 Sep · Salary+ 84,500.00
05 Sep · Rent− 22,000.00
11 Sep · Transfer+ 1,50,000.00
18 Sep · Card− 6,240.00
Closing balance2,61,860.00
Signed by issuer—
Running checks0 / 6
Format & structure
Metadata & producer
Font & glyph consistency
Hidden layers & revisions
Digital signature
Arithmetic checks
Trust score—/ 100 Awaiting score
threshold 80
01Image files

Images & scans. Every pixel, questioned.

Photos, scans and screenshots of IDs, bills and certificates — checked for edits, generation and re-capture before you trust them.

Document vs. non-document
Metadata & EXIF analysis
Copy-move, splicing & pixel edits
Noise, compression & lighting
Re-scan and print-to-scan artefacts
Deepfake & GenAI detection
Deepfake · 0.91 copy-move
id-card.jpg2 regions flagged
Account statementPage 1 of 3
02 Sep · Salary credit1,20,000.00
05 Sep · Rent-35,000.00
09 Sep · Transfer in2,45,000.00
14 Sep · Card payment-8,140.00
Closing balance3,21,860.00
font mismatch · 2 glyphsincremental edit · rev 2signature broken
bank-statement.pdfproducer: “PDF Editor 11”
02PDF files

PDFs. Down to the last glyph.

Bank statements, invoices, payslips and digital certificates — inspected at the file-structure level, where most edits leave a trace.

Format & structure validation
Metadata & producer trail
Font, glyph & kerning consistency
Hidden layers & overlays
Incremental edits & revisions
Digital signature validity
03Content & identity

What it says, and whether it adds up.

Beyond the file itself — the numbers, the names and the identities are checked against each other and against the source.

Arithmetic checks on invoices & statements
Cross-field consistency
Cross-document matching
Third-party verification for IDs
Issuer & template matching
Duplicate & reused document detection
!Arithmeticopening + credits − debits ≠ closing (Δ ₹60,000)
!Cross-documentsalary on payslip ≠ statement credit
✓PAN verified at sourceABCPS1234K · name matches
✓Issuer templatelayout matches known issuer
claim-bundle · 3 documents4 checks
04How it works

From upload to verdict, in seconds.

01
Send a document
Upload in the Studio or call one API — images, scans and PDFs.
02
Agents run every check
Pixel, structure, content and identity checks run in parallel.
03
Evidence on the page
Each flag is boxed on the exact region, with the reason.
04
Score into your workflow
A trust score and verdict, sent by API or webhook.
05Backed by research

Kriyam TamperFlow. The benchmark behind the pipeline.

An open benchmark for document tampering detection under realistic processing pipelines — printing, scanning, photocopying and digital sharing. Models that look strong on pristine datasets often fail here; DocForensics v2 is built for exactly these conditions.

1,050 human-annotated authentic and tampered document images
5 state-of-the-art models evaluated — the best reaches a Region-F1 of only 0.10
Frequency-domain methods lose up to 68.7% of localization under photocopy
Kriyam TamperFlow: Benchmarking Document Tampering Detection under Realistic Document Processing Pipelines Avishek Jana¹ · Swati Kumari¹¹ Kriyam.ai Labs, GEOGO, India Abstract As Document AI systems increasingly automate document processing at scale across high-stakes industries, ensuring the integrity of every processed document is no longer feasible through human review alone. Even a single tampered field within a document can facilitate fraud, identity theft, or record manipulation before it is detected. In this paper, we introduce Kriyam TamperFlow, an open benchmark for evaluating document tampering detection under realistic document processing pipelines… zenodo.org/records/21972485Read paper →
06Where teams use it

Stop fraud at the first document.

For any team that relies on documents to make a decision — and answers to a regulator for it.

Lending & underwriting
Catch edited statements and inflated payslips before credit is approved.
bank statementspayslipsITRs
Insurance claims
Flag altered hospital bills and reused prescriptions at intake.
hospital billsdischarge summaries
KYC & onboarding
Spot tampered or AI-generated IDs and verify them at source.
AadhaarPANpassports
Procurement & AP
Check invoice totals, tax and vendor details before you pay.
invoicesPOsGRNs
07For developers

One endpoint. Every check.

Send a file, get back a trust score, a verdict and every flag with its page and coordinates. Use it on its own, or as a step in a Kriyam Flow agent.

# request
curl -X POST https://api.kriyam.ai/v1/forensics/analyze \
-H "Authorization: Bearer $KRIYAM_API_KEY" \
-F file=@statement.pdf
# response
{
"trust_score": 38,
"verdict": "possible_tampering",
"flags": [
{ "check": "font_consistency", "page": 1, "bbox": [402, 618, 520, 640] },
{ "check": "arithmetic", "page": 2, "detail": "closing ≠ sum" }
]
}
08What we build

Trusted AI for enterprises, where every decision is auditable.

Tell us about your documents and workflow, we'll help you to find the right solution with Kriyam.

Evidence on every decision

Every value, flag and verdict links back to its exact region on the page.

Deploy where your data lives

Kriyam Cloud, or inside your own AWS or Databricks — we’re official partners on both.

ISO/IEC 27001:2022

Certified information security management.

Aligned with DPDP & GDPR

Privacy by design for India and the world.

DO IT YOURSELF

Bring us your most suspicious document. See what we find.

Edited statements, AI-generated IDs, a total that doesn’t add up — drop it in the Studio and get a trust score with every flag marked, in seconds.