dorklab
Live demo Features How it works Use cases Developers Request early access
Built on Claude · Early access

Document intelligence you configure, not rent.

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.

sample-invoice.pdf
INVOICE
INV-2041
Northwind Traders Pvt Ltd
PO-77123
48,320.00
Extracted by Claude live
Invoice numberINV-2041
VendorNorthwind Traders
Total48,320.00
PO numberreviewPO-77123
3 fields accepted · 1 sent to a human reviewer
Document types your team can define
Track record

Our founder built a production IDP system that processed 450,000+ documents.

What broke at that scale, and what kept reviewers fast, shaped how Dorklab handles confidence, review and tuning.

97%
avg. accuracy
450,000+
documents processed
13.5M+
fields extracted
These figures come from a production system the founder previously built, not from Dorklab. Accuracy is the share of extracted fields that matched the verified value.
Live demo

You decide how sure is sure enough.

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.

Review below
Flow straight through
Waiting for a human
Prompt examples added
Each confirmed correction becomes a few-shot example for that field.
What you get

Everything an engineer used to hard-code, now a setting.

Template-free extraction

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.

Your document types, no code

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.

Confidence you control

Set a threshold per field and per document type. Anything below it is routed to a reviewer; everything above flows straight to export.

Review that teaches

Every correction is stored and fed back into per-field prompts and few-shot examples. Accuracy improves from real usage, with no model retraining.

Ask your documents

Open any document and ask in plain language. Claude answers from the document itself, so reviewers stop hunting through pages.

Built for sensitive data

KYC files and health records need care. Dorklab is designed around data minimisation, encryption, role-based access and a full audit trail.

How it works

From a blank screen to a learning pipeline.

01

Define

Create a document type and add its fields, data types and thresholds.

02

Extract

Upload one file or a whole batch. Claude returns every field with a confidence score.

03

Review

Low-confidence fields land in a review queue, side by side with the source page.

04

Improve

Corrections tune that field's prompt and examples, so the same miss happens less.

05

Export

Send validated data onward as JSON or CSV, or pull it through the API.

Use cases

Paperwork-heavy work, one configurable engine.

FINANCE

Invoices and payables

Totals, tax, vendors and PO numbers captured as a single flat document type, with review only where confidence is low.

ONBOARDING

KYC documents

Identity and address proofs read into clean records, with mismatches routed to a person instead of waved through.

HEALTHCARE

Health records

Lab reports and discharge summaries turned into structured fields, handled with strict access control and audit trails.

LEGAL

Contracts

Parties, dates, terms and renewal clauses pulled out, with plain-language questions for everything else.

For developers

Configure in the UI. Integrate with one call.

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.

illustrative API shape
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 }  ]}

Be one of our first design partners.

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.

Get in touch
dorklab
Dorklab is an independent product built on Anthropic's Claude API. © 2026 Dorklab.
sukant.jha@dorklab.tech