DocPrism combines deterministic forensic checks and practical analyst workflows so teams can move faster without lowering risk standards or compromising auditability.
Nine independent forensic engines run simultaneously. Results are normalized into a 0–100 risk score with per-signal evidence your reviewers can act on immediately.
Layer 01 — Critical
Character-level font analysis across every word. Edited documents leave subset font prefixes (ABCDEF+Arial) that reveal exactly which words were changed — with page and word coordinates as proof.
Layer 02 — Critical
Compares the visible customer name (OCR-rendered) against the original name in the raw PDF data layer. Catches name substitution attacks that completely bypass visual inspection.
Layer 03 — High
Reads raw binary structure to detect incremental edits, hidden revision layers, multiple %%EOF markers, and tampered cross-reference tables — proof the document was modified after creation.
Layer 04 — High
Checks creator, producer, and modification fields against 70+ known fraud tools — AI generators (Canva, Firefly, Midjourney), PDF editors (Acrobat, Smallpdf), and office suites. Stripped metadata is itself a signal.
Layer 05 — High
Extracts the full transaction table, verifies every running balance end-to-end, checks the closing balance, and applies Benford's Law to detect statistically improbable digit distributions in fabricated data.
Layer 06 — High
Semantic vector embeddings identify when a new document's transaction set closely matches one already on file — even when amounts, dates, or names have been changed. Catches recycled statements across applicants.
Layer 07 — Critical
Scans text for explicit fraud markers: sample watermarks, lorem ipsum filler, placeholder names (John Doe), template labels, demonstration disclaimers, and known fake document generator website references.
Layer 08 — Medium
Genuine bank statements have consistent column alignment, line spacing, and font usage. DocPrism detects alignment breaks, character spacing anomalies, and font switching mid-line — artifacts of desktop editing.
Layer 09 — Configurable
A team-managed pattern library. Flag known fraud tool producers, match transaction signatures to prior cases, and blacklist document fingerprints. New patterns added by admin — no code changes needed.
Not a lab prototype. Built to support the daily throughput of underwriting, compliance, and servicing operations.
No. It accelerates and standardises review by surfacing high-confidence forensic signals and risk context, so analysts spend time on meaningful decisions rather than manually inspecting clean documents.
Bank statements, pay stubs, W-2, 1099, P60, P45, SA302, and PAYG summaries. PDF, JPG, PNG, TIFF, and WebP formats. US, UK, and AU jurisdictions are supported with jurisdiction-aware analysis logic.
Yes. Institution records, negative database patterns, scoring thresholds, and review rules can all be configured by your administrator with no code changes required.
No. DocPrism is designed for private self-hosted deployment. Your applicant data never leaves your controlled infrastructure, and there are zero external API dependencies during analysis.
Signals across nine forensic categories are weighted and normalised to a 0–100 scale. Below 30 is Low; 30–60 Medium; 60–80 High; 80+ Critical. Each category contributes independently so you can see exactly which signals drove the score.
We can run DocPrism against your sample statement types and walk through the forensic output with your team — no commitment required.
