Skip to content

AI Document Review

Evidence review that reads the document, not just the filename.

Upload a document to a matter and get back structured, reviewable facts — classified, OCR'd where needed, and extracted into the matter's record only after a human approves it. Here is the real, five-stage pipeline behind that.

The Pipeline

Five stages, evidence to reviewable facts.

Upload

A document is added to a matter — a filing, a supporting exhibit, correspondence, or any other piece of evidence.

Classification

The document is examined and a document type is proposed automatically, so downstream extraction knows what kind of facts to look for.

Text extraction & OCR

Text-based PDFs are read directly. Scanned or image-based pages — the common case for evidence that was originally paper — run through OCR to produce machine-readable text.

Structured fact extraction

Type-specific extraction rules pull discrete, structured facts out of the document text — not a summary, but the actual data points a matter record needs.

Human approval gate

Extracted facts are presented for review. Nothing is written into the matter's permanent record until a human approves it — the extraction step proposes; it never silently commits.

Why This Is Harder Than It Sounds

Most immigration evidence didn't start as clean text.

A meaningful share of the documents that matter in an immigration case — scanned passports, faxed employment letters, photographed certificates — have no underlying text layer at all. Running OCR client-side, in the browser, means a document's image never has to leave the browser just to become readable text.

Text-based PDFs skip the OCR step entirely and go straight to extraction — no reason to run image recognition on a document that already has a text layer. The pipeline picks the right path per document rather than treating everything as an image.

After Extraction

Facts become part of the grounded record.

Once a human approves what was extracted, those facts become part of the matter's record — the same record grounded drafting and case strategy draw from. A fact that was never approved never enters the record, and nothing downstream can cite it. For the retrieval and citation-verification mechanism this feeds into, see How XYRA's Immigration Intelligence Works.

See it work on a real document.

Private Alpha is open to a limited number of immigration firms.

Request Private Alpha Access