OCR ensemble · confidence scoring · human-in-the-loop
Paper forms in.
Clean records out.
SmartForms reads scanned documents with a dual OCR engine, extracts every field with a confidence score, and asks a person only about the fields it isn't sure of. Recurring forms learn a template and stop needing review at all.
- printed-text accuracy
- 95%+
- printed-text accuracy
- documents per batch
- 50
- documents per batch
- review pass, then zero-touch
- 1
- review pass, then zero-touch
A pipeline, not a black box.
Upload anything on paper
Scans, photos, faxes — PDF, PNG, JPEG, TIFF. Drop 50 documents in one batch, or send them straight to the API.
Two OCR engines read it
A Tesseract + PaddleOCR ensemble reads printed text; layout analysis finds the fields, tables, checkboxes and barcodes.
Every field gets a score
Extraction is confidence-scored per field. Above the threshold, it's done. Below it, the field is flagged — never silently guessed.
People review only the flags
The review screen shows the original snippet beside the extracted value. Fix the uncertain fields, approve, move on.
Records flow out
Export JSON or CSV, receive signed webhooks, or query the REST API. Recurring form types match a saved template and skip review entirely.
Everything between the scanner and your system of record.
Dual-engine OCR
Tesseract 5 + PaddleOCR read every page, in 50+ languages, with image cleanup (deskew, denoise) first.
Confidence-scored review
Per-field scores route only uncertain values to a human. The rest never wait in a queue.
Smart Import for spreadsheets
Upload a CSV or Excel file and get a proposed column mapping with evidence for every match — approve it once.
Templates that learn
An approved mapping is saved. The next file with the same shape maps itself; only business-critical fields still ask.
Dynamic datasets
Define record types from your own data — no schema migrations. Fields, types and validation live in a registry you edit.
Gap forms
Rows that fail validation generate a one-shot form containing exactly the missing fields. The answers repair the import.
Data classification built in
Columns are classified (PII, PHI, payment data) before anything else happens. Card numbers can never appear on a public form.
API, webhooks, exports
REST API with keys, HMAC-signed webhooks on completion, JSON and CSV export. Staging data purges itself on a retention clock.
Healthcare
patient intake · claims · referrals
Finance
applications · invoices · KYC forms
Legal
case intake · discovery · court forms
Government
permits · benefits · public records
Insurance
claims · FNOL · policy changes
Logistics
PODs · bills of lading · manifests
Priced by pages, not by seats.
Starter
Prove it on your own documents.
- 100 pages / month
- Single-document upload
- Confidence-scored review
- JSON & CSV export
Professional
Most teamsFor teams processing daily paper.
- 10,000 pages / month
- Batch upload (50 docs)
- Smart Import + saved templates
- Dynamic datasets & gap forms
- API keys & signed webhooks
Enterprise
Compliance-heavy volume work.
- Unlimited pages
- SSO & role-based access
- Data classification & retention controls
- On-prem deployment option
- Priority support
Asked before signing.
How accurate is the extraction?
Printed text targets 95%+ accuracy through a dual-engine OCR ensemble. More importantly, every field carries its own confidence score — anything below the threshold is routed to a person instead of being silently guessed, so low-confidence values never enter your data unreviewed.
What file types can I upload?
PDF, PNG, JPEG, TIFF and BMP for scanned documents — up to 50 in one batch. Smart Import additionally accepts CSV and Excel (XLSX) files for spreadsheet-to-record mapping.
What happens to sensitive data like SSNs or card numbers?
Columns are classified before anything else touches them — PII, PHI and payment data are detected with checksums, not just patterns. Payment-card fields can never be exposed on a public form, staged data is deleted on a retention clock, and webhooks are HMAC-signed.
Do I have to re-map the same spreadsheet every month?
No. The first import is reviewed once; the approved mapping is saved as a template. The next file with the same shape maps itself automatically — only business-critical fields (IDs, amounts, statuses) still ask for a confirmation.
Can it feed my existing systems?
Yes — JSON and CSV export, a full REST API with API keys, and signed webhooks that fire the moment processing or an import completes.
Is there an on-premises option?
Enterprise plans can deploy the full stack on your own infrastructure — it ships as Docker Compose with a Helm chart for Kubernetes.
Final section — signature required
The last time anyone retypes that form.
Upload a real document from your backlog and watch it become a record. Free tier, no card required.