White papers

White papers published here detail some of our projects involving the AI and Scientific Computing expertise applied to improve our client's business processes.

Natural Language Processing: looking for Value in your Unstructured Data


A frequent business problem is discussed, whereby companies find themselves facing a large base of Unstructured Data, which proves cumbersome to process. A specific case is brought up in which the data arrives continuously from the users and is processed on the fly. A comprehensive approach to processing this Unstructured Data is discussed along with technical considerations.

Keywords: #ai #artificialintelligence #ml #machinelearning #python #nlp #csv #gpu #gcp #jupyter #docker #tensorflow

Problem statement

We seek an approach to handling Unstructured Data on the fly with Machine Learning techniques, particularly with Natural Language Processing.

Creditworthiness Assessment: is your Client Creditworthy (and how do you make them so)?


In this white paper we discuss building a piece of software which performs Creditworthiness Assessment based on hundreds of parameters included in loan applications for a popular motorcycle brand. The process involves creating an algorithm which calculates the risk associated with each loan. The goal is to suggest modifications to the loan application parameters, which would adjust them so that they represent good risk for the expected return.

Keywords: #python #optimization #lp #linearprogramming #credit #creditworthiness #calculation #numpy #scipy #sympy

Problem statement

As a creditor issuing loans for a popular motorcycle brand, we want to understand, whether a particular loan application represents good risk to us and, if not, how we can revert to the applicant with recommendations on how they can modify their application to pass our creditworthiness criteria.

Fraud in Health Insurance: how do you detect it with Machine Learning?


In this white paper we depict a problem occurring in Health Insurance, namely fraudulent claims. The challenge lies in sieving them out while retaining the legitimate ones. There are plenty of potential techniques for Fraud Detection, ranging from Supervised Learning to Unsupervised Learning. Due to the availability of abundant training data we decide to go with Supervised Learning in general and Deep Learning in particular. A number of technical considerations are discussed.

Keywords: #ai #artificialintelligence #ml #machinelearning #python #tensorflow #fraud #frauddetection #deeplearning #dnn

Problem statement

We want to define and implement an AI/ML-enriched process, which would enable us to detect fraudulent insurance claims and highlight them to relevant parties, who can take appropriate actions on them.

Want more? Get in touch!