Document intelligence · Classify · Extract · Translate

Every document, read into clean structured data.

Anvaya AI reads the files Indian financial services and healthcare actually receive — scanned invoices, bank statements, KYC documents, handwritten doctor notes, medical bills, vernacular claim files — and returns typed fields with a confidence score and the exact place on the page each value came from.

Request access → See the API

every value, boxed where it
sits on the page

TAX INVOICE

vendor_nameSharma MotorsGSTIN 27AABCS1429B1ZK

Invoice No.invoice_numberINV-2041

Dateinvoice_date12/02/2026

Billed toAxis Claims Dept.

Denting — front panel18,500.00

Windshield22,400.00

Labour6,200.00

CGST + SGST (18%)8,478.00

Grand Totaltotal_amount₹ 55,578.00

E. & O. E. — Authorized signatory

POST /v1/extract · 200 OK · 9.4s

Field	Value	Confidence
invoice_number	INV-2041	0.98
invoice_date	12/02/2026	0.98
vendor_name	Sharma Motors	0.99
vendor_gstin	—not printed on this invoice	null
buyer_name	Axis Claims Dept.	0.95
tax_amount	₹ 8,478.00	0.97
total_amount	₹ 55,578.00	0.95

{
  "invoice_number": {
    "value": "INV-2041",
    "confidence": 0.98,
    "box": { "page": 1, "x": 0.66, "y": 0.24 }
  },
  "vendor_name": {
    "value": "Sharma Motors",
    "confidence": 0.99
  },
  "total_amount": {
    "value": "₹ 55,578.00",
    "confidence": 0.95
  },
  "usage": { "pages": 1, "est_cost_usd": 0.013 }
}

3-in-1

classify · extract · translate

1 = 1

one page, one credit — nothing else

ap-south-1

India residency, encrypted at rest

30 days

automatic document purge

Three operations

One API for the whole document pile.

Send a PDF or an image. Small documents answer synchronously; big ones run as jobs with webhooks. Every operation meters in pages, nothing else.

/v1/classify

Classify

Split a combined file into its logical documents and label each — with your own type list, or the model's.

{ "doc_type": "discharge_summary",
  "pages": "1–3", "language": "English" }
{ "doc_type": "doctor_notes",
  "pages": "4–4", "language": "Marathi" }

/v1/extract

Extract

Define fields in plain language — or let the model choose — and get values, confidence scores and bounding boxes back.

{ "name": "total_amount",
  "description": "grand total payable",
  "type": "amount" }

/v1/translate

Translate

Faithful, complete English translations of vernacular pages — handwriting included, never summarized.

{ "page": 2, "language": "Marathi",
  "html": "<h2>Witness Statement</h2>…" }

Built for Indian financial services

The unglamorous parts, handled.

Vernacular and handwritten pages Hindi, Marathi, Tamil and more — including handwritten doctor notes and witness statements — read, extracted and translated without dropping a line.

Every value carries provenance A confidence score, a reason when confidence is low, and the rectangle on the page it came from.

Data stays in Mumbai Hosted in AWS ap-south-1 on encrypted storage; documents purge automatically after 30 days.

Simple page-based pricing One page processed is one credit. No per-field or per-model arithmetic.

Sideways scans fixed first A deterministic orientation pass squares up rotated pages before any model reads them.

A console for the whole team Projects from templates, usage metering, API keys and user management ship with every account.

The API

A curl away.

# extract three fields from an invoice
curl -X POST https://api.anvayai.com/v1/extract \
  -H "Authorization: Bearer $ANVAYA_KEY" \
  -F "file=@invoice.pdf" \
  -F 'schema=[
    {"name":"invoice_number","description":"the invoice number"},
    {"name":"invoice_date","type":"date"},
    {"name":"total_amount","type":"amount"}
  ]'

Templates for invoices, bank statements, medical bills, discharge summaries, Aadhaar and driving licences come built in. Write to us for a key and a walkthrough.