Lattice

raylab.cloud

From a stack of approved PDFs to a register you can defend, tag by tag.

Register-driven · Evidence-backed · Auditable
The pipeline, end to end
01

Extract

Reads cached PDFs with Azure Document Intelligence and GPT-4o, recovers the tags the model misses, and normalises every one to a single canonical form.

02

Validate

Measures everything against the Master Tag Register, the yardstick. Coverage is counted, not assumed: exact match, punctuation-blind match, or flagged as possibly OCR-garbled.

03no AI cost

Audit

A read-only toolset over the cached documents, showing where they agree, differ only in wording, or genuinely contradict one another about the same tag.

04

Consolidate

One row per register tag: values resolved from the rows and from document title blocks, a queue for genuine conflicts, and an evidence trail behind every value.

Why it holds up

The register is the identity

Documents fill it in. They do not define it.

Triage, not resolution

Genuine disagreements stay blank and go to a human queue. It never invents a winner.

Honesty is the product

Every count states its caveats.

Every value is traceable

Each resolved value carries a trail back to the document it came from.

Every tag, in its place. Open n8n →