Introducing CogniDocs

The documents machines could never read.
Now they can.

CogniDocs reads the world's hardest documents — engineering drawings, bank and card statements, and legal records such as contracts, leases, policies, and case files — then turns them into structured, evidence-linked data a human can verify.

Read. Itemize. Classify. Prove it.

Finance parsing is available in the workspace today. Engineering results below come from scoped engagements; broader Engineering workspace access and Legal workflows are planned. Legal output is illustrative, not legal advice, and remains for counsel to verify.

Book a demo → See how ↓
COGNIDOCS · BANK STATEMENT — MERCHANT NORMALIZATION

Bank statement example

RIDEHUB*TRIP 8842RideHubUSD 34.00
SHOPMART*ORDER 2K4LP0QW3ShopMartUSD 187.25
QUICKBITE*ORDER 654321QuickBiteUSD 62.40
FUELWAY*PUMP 715FuelWayUSD 120.00
STREAMBOX*MEMBERSHIP 4099StreamBoxUSD 39.00
FRESHMARKET*GROCERIES 441FreshMarketUSD 254.80
SKYWAYS*ONLINE BOOKINGSkyWaysUSD 2,340.00

FORMAT LEARNED · NEXT STATEMENT: 2.8 SECONDS · 214 LINES, ALL CATEGORIZED

Floor plan example

47 ITEMS READ

MAJLIS
15.9 m² ✓LIVING
18.6 m² ✓KITCHEN
5.2 m² ✓BED 1
2.9 m ✗

EVERY FINDING PINNED TO SHEET & ROOM · CLAUSE QUOTED · REVIEWER DECIDES

Select an example to explore the demonstration.

· DWG· DXF· VECTOR PDF· IFC / BIM· P&ID· SCANS· BANK STATEMENTS· CARD STATEMENTS· INVOICES· BLUEPRINTS· CONTRACTS· LEASES· POLICIES

Act I — The blind spot

The world's money, infrastructure, and obligations run on documents that were never written for machines.

Drawings are drawn, not written

A floor plan's most important facts aren't text — they're lines, symbols and dimensions. Labels get drawn as outlines. Scale bars lie. OCR sees pixels; the meaning lives in the geometry.

Statements speak in cipher

Every bank, a different format. Every line, a cryptic string — “CRM*CAREEM RIDES DXB 8842” means nothing to a ledger. Multiply by thousands of templates that change without notice.

Contracts carry meaning across clauses

Parties, dates, duties, rights, renewals, governing law, exceptions, schedules, exhibits, and amendments require context and source evidence — not keyword search alone.

So people re-type

Analysts, engineers, and legal teams transcribing, abstracting, reconciling, and re-checking — slow, expensive, and exactly where errors hide. The data was always there; nobody could read it.

Act II — The breakthrough

One reading engine. Three layers deep.

No single technique can read these documents — so CogniDocs doesn't use one. Three extraction layers read every document independently, then fuse: what one layer misses, another catches. That fusion is the IP — and it's what pushes accuracy past 90% at production speed.

And it refuses to guess. AI finds things; deterministic code does every measurement, comparison and sum. Where the evidence is insufficient, CogniDocs says so — every extracted item carries its source and its confidence.

ONE COORDINATE SPACE

3

Computer vision

sees stamps, scans, handwriting — what isn’t text and isn’t vector

2

Text & coordinates

every character extracted with its exact position: sheet, room, cell

1

Vector geometry

the drawn lines, symbols & tables themselves, read as structured data

Fused together: 90%+ accuracy — with every item traceable to exactly where it was found.

One engine, three editions

Wherever documents hold the numbers, CogniDocs gets them out.

CogniDocs Finance

Statements become structured, categorized data.

Bank and card statements from any institution, itemized line by line — with merchant names normalized against a global database of hundreds of thousands of brands and categorized by industry.

  • ✓Merchant intelligence — mangled descriptors resolve to the right global brand and industry.
  • ✓Self-learning formats — the first statement teaches the engine; every one after processes in seconds.
  • ✓Every line auditable — amount, date, merchant, category, and exact source location.
ReconciliationLending & underwritingExpense managementAccounting data entrySpend analyticsAudit tie-outs
CogniDocs Engineering · Engagement-proven

Drawings become itemized, checkable bills of fact.

Every discipline of a drawing pack — architectural, structural, mechanical, electrical, energy efficiency — plus pipeline and oil-installation drawings and BIM models. Read natively (DWG · DXF · vector PDF · IFC), itemized at a 90%+ hit rate, and pinned to sheet and location.

  • ✓The whole pack, all disciplines — architectural, structural, mechanical and electrical sheets in one pass.
  • ✓Native CAD reading — dimensions and layers read as data, never guessed from an image.
  • ✓Evidence-pinned findings — every quantity points to the exact sheet and spot.
  • ✓On-premise ready — runs offline on ordinary hardware.
BOQ & quantity takeoffCode-compliance reviewTender & bid evaluationPermit reviewP&ID digitization
CogniDocs Legal · Planned

Legal documents become clause-level, evidence-linked records.

Legal departments face rising matter volumes with constrained teams, while obligations and risk-critical terms remain buried across contracts, leases, policies, and case files. The planned CogniDocs Legal workflow would read them clause by clause — structuring parties, dates, duties, rights, renewals, governing law, and deviations, with every extraction pinned to the exact provision for counsel to verify.

  • ✓Clause and obligation extraction — would flag and structure parties, dates, duties, rights, renewals, governing law, and deviations for counsel to verify.
  • ✓Playbook-based review — would compare material with your approved playbook and flag deviations for counsel to verify, not final risk decisions.
  • ✓Whole-matter context — would connect agreements, schedules, exhibits, and amendments so counsel can review the complete record.
  • ✓Evidence-pinned findings — would carry page, clause, and confidence so counsel can verify every source.
contract abstraction & reviewM&A due diligenceCLM migration & data cleanupobligation managementrenewal & expiry monitoringNDA triagelease abstractionprocurement contract reviewpolicy compliance mappinglitigation chronologye-discovery document triagecase-file indexing

Proven on real work, not benchmarks

Real permit drawings. Real RFQ packs. Real statements.

These Finance and Engineering figures come from live engagements — measured on actual client documents, re-verified by built-in regression tests, reproducible on demand.

878

printed dimensions read exactly from real permit drawings — zero guessed

80

drawing sheets read across five disciplines

Days → 1 hr

a full RFQ pack turned into a priced, cross-checked BOQ the engineer only reviews

100%

can run on-premise, offline, CPU-only — documents never leave the building

Every figure carries its source. A number without evidence is not a number.

The Legal contract workflow shown above is illustrative and planned; it is not represented by these production metrics and remains subject to counsel’s source verification.

Better together

CogniDocs reads it. CogniOS runs it.

Alone, CogniDocs is a document-intelligence engine with an API — drop it into your existing systems. Together with CogniOS, it becomes the sense layer of the agentic organization.

CogniDocs

Itemizes tender bids, reads permit drawings, classifies statements, and illustrates planned contract review — handing over structured, evidence-pinned data for human verification.

CogniOS

Routes that data through governed workflows — bid evaluation, permit review, reconciliation, and planned contract review, obligations, and renewals — where agents prepare and a human decides.

One planned example: a complex contract arrives → CogniDocs illustrates source-pinned obligations → CogniOS prepares the renewal workflow → your counsel verifies it.

Explore CogniOS →

The standing challenge

Bring us your hardest document.

A drawing pack, a scanned statement, or a complex contract. We’ll show structured, source-pinned review in front of you — with planned Legal work clearly illustrative and a human in control.

Book a demo →