Technology: What’s under the hood?

How we’ve trained our AI to fully understand documents

Breakthrough computer vision and natural language processing for document understanding

Pre-trained from 55 million industrial documents, including invoices, receipts, packing lists, shipping labels, and forms

The pre-trained AI requires no additional training when deployed. Deployments can be as fast as 1 day.

Supercharge your document workflows today

656167_Make data graphic_V1_022620.png
logistics 1.png

Photon provides a universal data capture service to breakthrough deep learning, image classification, object recognition, image processing, and natural-language processing algorithms under the hood.

Photon's proprietary machine learning models have been trained on the largest datasets of industrial documents in the world.

​Its training gets smarter over time. It returns structured data from any image, document, or scan by understanding the type, spatial layout, format, and data types of each field.

Read the API Documentation

XMLID 953.png

Computer vision pipeline

  1. ​Image quality triaging

  2. Automatic thresholding

  3. Boundary detection

  4. Cropping regions of interest

  5. Document classification

  6. Structured layout comprehension

  7. Object detection

  8. Barcode scanning

  9. Image transformations, auto-rotations

  10. De-skew, affine transformations

  11. Printed vs handwritten text classification

  12. ​Zonal analysis

  13. Cascaded, semi-supervised deep learning model

  14. Pooling

  15. Ensembling​​

  16. Text capture

  17. Signature validation

  18. Cascaded case escalation

  19. Structured text and barcode data output

aigear 1.png

 Natural language processing pipeline

  1. Constituency parsing

  2. Semantic parsing

  3. Bidirectional Recurrent Neural Networks (LSTMs)

  4. Word vector space model similarity matching

  5. Structured field matching

  6. Data types validation

  7. Domain-specific dictionary lookups

  8. Spell checking

  9. Selective case escalation

  10. Relation extraction

  11. Name matching

  12. Address validation

  13. Tracking # lookups

  14. UPC, SKU, item lookups

  15. Invoice, PO, receivables, ASN matching

  16. Database queries

  17. Exception handling and reconciliation

  18. Data types transformations, Extract Transform Load

  19. API integrations with WMS, ERP, IMS, DB

 The Intelligent Document Processing solution for modern businesses

  1. Pre-extraction:
    Performs image pre-processing to increase the quality of the scanned document, captures, data, and indexes and classifies the documents into categories

  2. Extraction:
    Captures relevant data leveraging Natural Language Processing for further processing

  3. Post-extraction:
    ​Validates the extracted data with the help business logic, validation rules, and enterprise databases