DEPS Document processing, data extraction software

Problem Statement

Our clients' manual process of extracting data points in different formats and from different vendors was time consuming and prone to error.

Solution Proposed

• A single .pdf file can contain multiple DDRs, which need to be split.
• Vendors all have their own DDR formats, layouts, and field names that must be reconciled.
• Even a single vendor may have several different formats.

Achieved Results

• Train an ML-based model to find and extract required data in natural language.
• Recognize synonyms to unify extracted data across all vendor formats.
• Evaluate the confidence level for each extraction field.
• Validate extraction data with business rules and submit it automatically.
• Export output data in both automation and human-readable formats.

Problem Statement

Our client, a large law firm, was looking for a reliable, cost-effective way to extract and digitize data from various document types (judgments, tax records, liens, etc.) to use in compiling orders for their clients. Extracting data manually is costly, error-prone, and tedious.

Solution Proposed

• Over 50 distinct types of documents varied by geography and age.
• Over 50 distinct types of documents varied by geography and age.
• Documents provided in compound packages requiring splitting and classification.

Achieved Results

• Custom ML models that automatically extract data.
• A classification model to define document type.
• Optimized use of a paid OCR engine.

Problem Statement

Our client was plagued with the time-consuming and labor-intensive task of searching and verifying RFP submissions for specific fields across various file formats.

Solution Proposed

• The changeable nature of editable PDFs.
• The need to reconcile PDF, XLS, and XLSX files from multiple vendors with different document versions.
• Numerous regulations and exceptions to manage.

Achieved Results

Our efficient and reliable solution trained a custom ML model and a classification algorithm for RFP document versions that:

• Automates data validation aligned with business rules and a defined data dictionary.
• Accurately extracts data from editable PDFs, including changing field positions.
• Provides flexible field management.

Problem Statement

A large volume of scanned documents (PDF files) in hundreds of formats ⠀ ⠀

Solution Proposed

• Advanced ML techniques (bidirectional recurrent neural networks) on top of the processing pipeline packaging all extraction steps
• Supporting workflow for golden dataset creation (original document regions markup and preliminary data labeling)

Achieved Results

Digitalized three million documents and achieved a three-times faster turnaround with 90% accuracy

Problem Statement

Process photographed receipts to collect and analyze the list of bought items

Solution Proposed

• Custom ML model for automatic data extraction
• Identification of brand, product category, etc., based on item name through validation with client’s data storage
• Intuitive UI illustrating confidence level of item recognition
• Integration with customer services via API

Achieved Results

Opened a new market data stream by scanning 10,000 paychecks monthly across nine commercial networks and applying analytics solution

Problem Statement

Multilanguage 200+ pages documents processed by humans, requiring up to 10 hours to handle one document

Solution Proposed

• Custom ML model for table identification, data search and extraction

• Automatic text translation from various languages to English (integration with a cloud-based translation engine)

• Rule-based value validation highlighting mismatched fields

• Intuitive UI for easy document management to eliminate work duplication

Achieved Results

• Data is searchable across all documents including tables

• 40% faster processing and quicker time to market

Problem Statement

5,000 to 10,000 pages per document with multiple tables of complex structure

Solution Proposed

• Integration of several cloud-based solutions and client’s proprietary system into a single pipeline for cost optimized and high quality extraction

• Intuitive UI for easy document management to eliminate work duplication

• Check-box recognition service

Achieved Results

• Data is searchable across all documents including tables

• Processing of multiple 10,000-page documents

• Reduced business cost for manual data processing unit by 70%

• Three integrated and cost-optimized services

Problem Statement

• Effort intensive and error prone processing of application forms

• Input data in different formats (image, PDF (searchable and unsearchable, fax, docx, or a combination), printed and handwritten

Solution Proposed

• Enabled robotic email processing and attachment extraction with following document classification per format (for further processing of PDF and docx)

• Automatic rule-based prioritization and intake of file batches

• Documents are pre-processed for image quality improvement (blurring, smoothing binarization, etc.)

• Automated classification per vendor who use different templates

• Applied OCR – Tesseract, OMR (Optical Mark Recognition) and ICR (Intelligent Characters Recognition) and Google Vision API - for extraction of data ranging in quality

• Enabled post-processing for maximum accuracy: data correction (type-based validation), cross-field validation (required fields should be completed) and check box validation across document

• Data search (powered by InfoNgen)

• Advanced data search via AI text analysis

Achieved Results

• Optimized process workflow with quick and stable document intake

• Faster processing and higher operation accuracy

• 4,000 forms processed monthly

• OCR, OMR, ICR used for low quality documents

Problem Statement

Manual processing of 100,000 personal documents per year, each applying 100+ validation steps manually per document

Solution Proposed

• ML-solution for personal document recognition

• AI service for quality check of uploaded data that requests user to retape photos with shadows and flares

• Automated document validation

• Customization of validation steps

• Intuitive UI for easy document management to eliminate work duplication

Achieved Results

• Reduced business cost for person validation

• 80% faster processing and quicker user application processing

• 100,000 forms per year with tables

• High sensitivity data processing

Problem Statement

Client needed to load high-quality property-related data into their database and reduce errors with the real estate data exchange process

Solution Proposed

• Rule-based value validation with highlighting mismatched fields
• Ability to update business rules from UI without code changes
• Intuitive UI for easy document management
• Support of different user roles and groups
• Algorithm for defining connections between property data implemented from scratch

Achieved Results

• Data integrity guaranteed by validation rules
• Convenient tree-like property data structure