Use Case Portfolio

Perform clinical note extraction from EMR/EHR charts and store information within a clinical schema for natural language processing (NLP). NLP design will consist of a framework for disease prediction that combines both clinical notes and structured data. Our approach will also explore the idea of a design NOT requiring disease specific feature engineering and allow for handling of negations and numerical values within the notes extracted.

Using Machine Learning, automatically extract vitals such as temperature, blood pressure etcetera from  medical records that can be used downstream to help care management operations. The approach taken relies on a semantic lexicon and extraction rules strategy. 

Making better credit lending decisions for current and potential borrowers is highly important to a lender. Our model determines a borrowers credit worthiness and propensity to default. We applied various models  to evaluate and validate metric values, accuracy, precision, recall and f1 score for our benchmark (Logistic Regression), Random Forest and Gradient Boost models. Each of our models have high accuracy scores and can be designed to customer needs.

Under Construction

Orientation correction has been a long standing task in document analysis. Most methods exploit the special structure of document images, such as text layout in lines and precise shapes of letters. In general, the task is harder since important features such as text or picture boundaries are not available and even image features such as the horizon or other dominant horizontal or vertical lines in the scene can be missing. The approach taken to solve the proper orientation for faxes, while simple, allows for orientation of 'any' type of image, learnt thru a supervised training approach with heavy emphasis on transfer learning (domain adaptation). Also, within this approach a lessor known method for tuning one of the dominant hyper-parameters is employed (Cyclical Learning Rate).

Using supervised learning, identify members with type I and type II diabetes and then predict the likely-hood that a member will be re-admitted to the hospital within 30 days of being released. 

Under Construction

Under Construction

Under Construction

Under Construction

Employed several supervised algorithms to accurately model individuals' income using data collected from U.S. Census. Identified the best ML algorithm from preliminary results and further optimized this algorithm to best model the data. Criteria for implementation was to construct a model that accurately predicts whether an individual makes more than $50,000. This sort of task can arise in a non-profit setting, where organizations survive on donations. Understanding an individual's income can help a non-profit better understand how large of a donation to request, or whether or not they should reach out to the individual. While it can be difficult to determine an individual's general income bracket directly from public sources, we can infer this value from other publicly available supplemental data.

Our custom solution is an interactive digital assistant that can take advantage of publicly available information about a company and provide that to an end user through voice commands. Our models cognitive functions provide the ability to learn information from the use of spiders' (a way of collecting information from www) and also from supervised modeling. Cognitive functions also include identification of team members through voice recognition, establishing security access through other bio-metric information and the ability to behave as a greeter.

Developed a framework for fast tracking model development with hyper-parameter automated tuning. The code base will be constantly evolving and enhancements added based on need. With that stated, the current version allows the user to effectively train up to three supervised models at once and auto-tune hyper-parameters for each of the models selected. Although this ability can be easily expanded to support more, it is not recommended if using on a CPU only system. Planned future enhancements include un-supervised and transfer learning along with the ability to add one’s own model architecture.

And Many More...….