Carlos del Cacho

Crawling of social media (Twitter) and filtering based on specific keywords to retrieve and classify sentiment with regards to the Peace Plebiscite that took place in Colombia in October 2016. Classification through recurrent neural networks with deeplearning4j.

Developed a personalization module based on ElasticSearch and text classification through linear kernel SVMs that attained 93% accuracy in tagging video metadata content automatically. Lead the development team and managed customer engagement. The system is designed over cloud technology in AWS to withstand peak traffic loads of over 150 million page views per month for the Tokyo 2020 games.

Developed a hierarchical logistic model tree for assessing loan risk that departed from the bank's methodology and incorporated non-traditional signals from public datasets such as events in the Companies House Records (changes in management, refinancing, bankruptcies, etc).

Created a distributed system to predict daily unit product sales in over a thousand stores across Spain, performing 20 million regressions on a daily basis, improving accuracy, and delivering cost efficiencies in the supply chain by reducing excess inventory while avoiding stock-outs. Technologies: Apache Spark, Python, and scikit-learn.

Developed an internal software with custom ranking functions over Apache Solr to dynamically check for article inventory in over a thousand stores with response times in milliseconds, prioritizing available locations by distance in search engine results.