DeepFusion/Vision API

Our Vision API and NLP based Deep learning algorithms along with NER capable contextual Smart Analysis Platform “DeepFusion” enables highly accurate submission pipeline processing, policy servicing, claims settlement, while extracting foresightful analytics from data that is otherwise overlooked.

DeepFusion has Pre-Trained BFSI domain specific ML Models to classify documents, Cluster documents, Segment images, Segment Texts, Text sentiments, Custom NER etc. and these models can be extended and customized with specific needs. Models are Ensemble for better accuracy and provide Explainable AI to understand the decision in order to create trust and for auditing.

With DeepFusion,
with 100% automation, you can

Automatically classify documents

Detect signature/photo in the document and recognize the owner of the document

Detect a document containing the expected contract clause and validate the correctness

Analyze document layout, avail image/text recognition and verification

Report for a particular text/data in a document

Cluster and organize documents on similarity and/or category basis

Derive the sentiment of the document

Detect and recognize logos, photos and images in the document

Summarize multi page document content

Link different documents

Extract key data from documents to validate against business rules

Verify the completeness of a document by analyzing document layout, using image/text recognition and verification

This is how DeepFusion improves operational efficiency and converts data into actionable insights

What Differentiates DeepFusion?

Image Preprocessing

Enhance images of low resolution and increase the accuracy of data detection and accuracy

Data Extraction Using
AI-based OCR

AI based OCR enables the following benefits,

1. Ability to locate objects and content with high accuracy irrespective of varying spatial locations,

2. Document layout analysis with zero-templates

3. Extraction of data from markups like checkbox, tables, drop-downs and radio buttons thereby benefitting data extraction not supported by common OCRs

Zero Rules Code For
Data Validation

Support for ML models to learn the rules by themselves during training phase,

ML models are frequently updated with new and unknown rules . This feature benefits clients on new rule detection and changing life cycle.

ML models dynamically make records linkage with external sources .

Support For Image Recognition And Verification

Deep learning models verify photos and hand-written signatures without manual intervention.

Common Models As APIs To Instantly Support New Use Cases

APIs and ML Models classify documents, cluster documents, segment images, and segment texts. Text sentiments, Custom NER, text similarity and signature verification are packaged as ML models. These models can be extended and customized with specific needs. These models present as API allow to quickly build business processes.

Threshold-based Incremental Learning Architecture

Every point in the workflow is managed through thresholds. Anything that does not meet thresholds is taken to exception queues, and with manual intervention and incremental learning capabilities, new rules and unknown patterns are instantly trained.