Multi-format ingestion
Contracts, policies, procedures and regulatory text from documents, scans and content systems.
AI intelligence system
ComplianceIQ
From document summary to document-to-document verification.
A verification engine that reads contracts, policies, procedures and regulatory text together and reports where they contradict each other, where evidence is missing and where an obligation has no controlling procedure.
- Sector
- Legal, Risk & Compliance
- Year
- 2025
- Duration
- 7 months
- Team
- 6 people
- Platforms
- Web · API
- Status
- In production
Claim-level
Unit of analysis
Not document-level
Both sides
Shown per finding
Passage against passage
4,000+
Documents in corpus
Reference deployment
Cited
Every finding
To document, clause and page
01 — The problem
The compliance function held four thousand documents. Contracts committed the business to controls that internal policy did not require. Policies referenced procedures that had been superseded. Nobody knew, because verifying it meant reading everything against everything.
Summarisation tools made this worse rather than better. A good summary of a bad policy is still a bad policy, and summarising each document separately is exactly the wrong operation when the problem lives between documents.
02 — What we did
We built for the space between documents. Obligations, controls and definitions are extracted as structured claims with their source location attached, then compared across the corpus. The unit of analysis is the claim, not the file.
Findings are asserted with both sides shown. A contradiction is reported as two passages side by side with the reasoning between them, so a compliance officer verifies in seconds instead of taking the system's word for it.
Summarising each document separately is the wrong operation when the problem is between them.
Interface
What it looks like in use
Two of the screens that carry the most weight in daily use, rebuilt here from the production design system.
Finding 882 · contradictionHIGH
MSA-2024-11 §7.3 · retain 7 yearsCLAUSE A
POL-DATA rev 9 §4 · purge at 24 monthsCLAUSE B
Both cannot hold for customer recordsCONFLICT
Owner · data governanceTRIAGED
Cited to page and clause
Both sides, side by sideA contradiction is presented as two passages with the conflict stated between them, so an officer verifies in seconds.
Framework coverage4,112 DOCS
Access control12 / 12
Data retention7 / 9
Incident notification · no controlling procedure0 / 3
74% requirements covered
Coverage stated, not assumedEach requirement shows the control that satisfies it, or shows plainly that nothing does.
Capabilities
What the system does
Grouped by the job each set of capabilities exists to do, rather than by which team built it.
Ingestion and extraction
01 · Making documents comparableClause extraction
Obligations, controls, definitions and exceptions extracted as structured claims with precise source anchors.
Definition resolution
Defined terms resolved per document, since the same word often means different things in two contracts.
Version awareness
Supersession chains tracked so comparison runs against the controlling revision.
Cross-document verification
02 · The core capabilityObligation mapping
Every external obligation mapped to the internal policy and procedure intended to satisfy it.
Contradiction detection
Claims that cannot both hold surfaced with both passages and the conflict stated explicitly.
Gap detection
Obligations with no controlling procedure, and controls with no evidence requirement attached.
Regulatory mapping
Framework requirements mapped onto the internal control set with coverage stated per requirement.
Review and reporting
03 · Getting it usedCitation-linked findings
Every finding opens directly onto the source passages on both sides.
Triage workflow
Findings routed by severity and owner with accept, dispute and remediate states.
Change impact
When a document changes, everything downstream that depended on it is re-verified and flagged.
Audit reporting
Coverage and exception reports generated from the live finding set.
Architecture
How it is put together
Layer by layer, with the reason each one exists — because the reason is usually the interesting part.
01Ingestion
Format-agnostic pipeline with layout-aware parsing and OCR for scanned material, preserving structure and page anchors.
- Layout parsing
- OCR
- Structure preservation
02Claim extraction
Structured extraction of obligations, controls and definitions with span-level source anchors.
- Extraction models
- Span anchoring
03Comparison engine
Candidate pair generation via semantic retrieval followed by entailment and contradiction classification.
- pgvector
- NLI models
04Verification
A validator requiring both supporting spans before any finding is published, discarding unsupported detections.
- Span validator
- Claude
05Workspace
Side-by-side review interface with triage, ownership and change-impact tracking.
- React
- PDF.js
- PostgreSQL
Technology
Document AI
- Layout parsing
- OCR
- NLI models
- pgvector
Backend
- Python
- FastAPI
- PostgreSQL
- Celery
Frontend
- React
- TypeScript
- PDF.js
Platform
- Kubernetes
- S3
- Audit logging
Outcome
What changed
Measured against how the operation ran before, not against a benchmark chosen after the fact.
-
Conflicts between documents became visible. Contradictions that had survived years of separate reviews were surfaced in the first full pass.
-
Coverage is stated, not assumed. Each regulatory requirement shows the control that satisfies it, or shows that none does.
-
Verification takes seconds per finding. Both passages are shown, so an officer confirms rather than investigates.
-
Document changes trigger re-verification. Downstream dependencies are re-checked automatically.
How it ran
-
Months 1-2
Corpus analysis
Assessed document types, formats and how badly supersession chains had degraded.
-
Months 3-4
Claim extraction
Structured extraction with span anchoring, validated against manually marked-up documents.
-
Month 5
Comparison engine
Retrieval-based pairing and contradiction classification with the span validator.
-
Months 6-7
Review workspace
Side-by-side interface, triage workflow and change-impact re-verification.
Related work
From the same practice
Manufacturing & Production Quality
LineTrace
From defect detection to defect causality.
A multimodal system that correlates line video, machine telemetry, batch records and operator actions to explain why a defect happened, not merely that it did..
2025 · 10 months
Telecommunications & Network Operations
NetCause
Not alarm detection. Failure hypothesis reduction.
A correlation engine that turns an alarm storm into a short ranked list of probable root causes by reasoning over network topology and fault propagation rather than alarm counts..
2025 · 8 months
Oil & Gas · Asset Integrity
PipeIntegrity
From a single inspection to longitudinal asset integrity.
A vision system that detects and localises pipeline deterioration, then tracks each defect across inspection cycles so maintenance is prioritised by progression rather than by appearance..
2024 · 9 months
Something similar?
Running into the same problem ComplianceIQ solved?
Tell us what you run. We will reply within two business days with what we would build for your situation — and, just as usefully, what we would leave out.