Dorklab turns invoices, KYC files, health records and contracts into validated, structured data using Claude. Your team defines the fields, sets the confidence bar and teaches the system from every correction, from a no-code interface.
What broke at that scale, and what kept reviewers fast, shaped how Dorklab handles confidence, review and tuning.
Pick a document, move the confidence bar and watch which fields flow straight through and which go to a reviewer. Confirm a flagged field and see the correction become a prompt example. This runs on sample data.
Point Dorklab at a scanned or digital PDF and Claude reads it the way a person would. No per-vendor templates, no layout rules to maintain.
Name a document type, list the fields, set data types and mark what is mandatory. A new type takes minutes, and an invoice stays one simple type, not a tree of sub-types.
Set a threshold per field and per document type. Anything below it is routed to a reviewer; everything above flows straight to export.
Every correction is stored and fed back into per-field prompts and few-shot examples. Accuracy improves from real usage, with no model retraining.
Open any document and ask in plain language. Claude answers from the document itself, so reviewers stop hunting through pages.
KYC files and health records need care. Dorklab is designed around data minimisation, encryption, role-based access and a full audit trail.
Create a document type and add its fields, data types and thresholds.
Upload one file or a whole batch. Claude returns every field with a confidence score.
Low-confidence fields land in a review queue, side by side with the source page.
Corrections tune that field's prompt and examples, so the same miss happens less.
Send validated data onward as JSON or CSV, or pull it through the API.
Totals, tax, vendors and PO numbers captured as a single flat document type, with review only where confidence is low.
Identity and address proofs read into clean records, with mismatches routed to a person instead of waved through.
Lab reports and discharge summaries turned into structured fields, handled with strict access control and audit trails.
Parties, dates, terms and renewal clauses pulled out, with plain-language questions for everything else.
Everything you set up in the interface is available over a simple API, so documents can arrive from your own systems and validated data can flow straight back.
POST /v1/documents{ "document_type": "kyc_identity", "file_url": "https://your-bucket/id.pdf", "review_threshold": 0.9} // response{ "status": "needs_review", "fields": [ { "name": "document_number", "value": "P1234567", "confidence": 0.89 } ]}
Tell us which documents slow your team down. We are working with a small group of early teams and shaping Dorklab around what they need.