Concord Capture

Machine Learning

NLP

Collaborative Learning System

Ability to process multiple formats of document/ images coming from various sources.
Extract certain data and patterns present in unstructured document/images.
Feature to add/ review new rules or data patterns for extracting definitive fields from faxed images.
Setup RPA pipeline to process incoming documents, extract all the relevant information across various stages.

Preprocessing images and performing image segmentation using Leptonica, Leptonica and TensorFlow to clean up unwanted sections/identify sections having possible textual contents.
The identified possible high value sections are fed to an OCR module, which extracts the raw text.
The text is fed to a dynamic rules based NLP Engine to extract the required data. The dynamic rules based NLP Engine gives us the ability to add rules on the fly without recomputing the rules model.
All these components are hooked up into an RPA pipeline. Each modules in the pipeline expose custom control APIs which is used by the RPA pipeline to monitor and control the flow of documents into each stage of the pipeline.
All individual modules were dockerised and deployed to production environment using Kubernetes. The RPA pipeline plays a role in the deployment orchestration and determines the number: of instances of each pipeline stage to handle and distribute the workload.