Become a Data Scientist

Details

Discrete Data Solutions provides software and data solutions to clients facing technology hurdles. In this instance, their client was a consulting firm with a large CRM, filled with thousands of outdated contacts. The client asked for a method to automate the updating of the contact information and addition of data from various social media sources.

In order to address this problem, Michael developed the following:

• Several automated web scraping tools
• Data processing pipeline that integrates data from multiple sources
• A series of data matching ML models that leverage word vectorization and Levenshtein or Mallow distance as well as several other methods
• Email format simulator based on the supervised learning models and trained with email addresses from others in the company and classified by when they joined the company

Tasks

• Web Scraping
• Data Cleaning and Processing
• Record Linkage
• Social Media Mining
• Email Format Prediction

Details

Parable reduces the cost & inconvenience of wound care through a platform that allows healthcare providers to better measure, monitor, and manage their wound patients. The platform facilitates care coordination and offers a mechanism to catch wound complications early. Specifically, Parable checks for clinical indicators of expected healing/development, captures a visual time lapse of the progression, and reminds patients (in ambulatory settings) about proper care instructions to ensure adherence.

For each case, a patient might check-in multiple times. We used data wrangling to capture the improvement of the wound between check-ins and later modeled that with the XGBoost implementation of gradient boosted trees to try to classify potential treatments either as an improvement or worsening of the wound, as well as to try to quantify what the improvement would be in an attempt to be able to recommend the best possible treatment for each case.

Tasks

• Data Wrangling
• Exploratory Data Analysis
• Problem Formulation
• Time Series Analysis
• Classification

Details

Havenly provides remote interior design solutions for users through their online platform. The design process includes product selection from the catalogs of partner retailers. Often the product classification provided is not consistent. This is due to many retailers maintaining their own hierarchy for product classification.

To address this problem, a process was developed for taxonomy classification using product descriptions provided by partner retailers. Natural language processing tools were used in conjunction with a support vector machine classifier to achieve an accuracy score of 0.99 on the test data set.

Tasks

• Natural Language Processing
• Taxonomy Classification