Edwin Silva Email & Phone Number

Colombia

Technical leader for fraud prevention. I was part a team working on fraud prevention, particularly ATO detection. Mercadolibre is a multibillion yearly revenue company and as such there are millions of USD in fraud per year. Using device (mobile and desktop) data, user behaviour, transactional data, geolocation and inferred information as features we built and updated an evolving machine learning model used to obtain a real time score of fraud likelihood to be used either directly to reject transactions, to ask for a stronger user reauthentication, to resend the transaction for manual review or to approve the transaction. Thanks to this and to other measures, in the period from May 2022 to December 2023 we were able to reduce the ATO impact by approximately 60% in quantity and 90% in TPV (this is, the fraud per transaction average was reduced, so the TPV impact was more dramatically improved).

Bogota,D.C., Capital District, Colombia

Casa Editorial El Tiempo is the largest newspaper from Colombia, with more than 100M page views and 10M unique users per month. My main role was the design and implementation of the automatic recommendation of personalized news content for registered users. The recommended news were displayed in the home page at eltiempo.com and it produced noticeable improved engagement metrics, such as 25% higher page views compared with the baseline (news manually put in a more prominent section of the home page by the editors). As a byproduct of the recommendation system, I also developed the personalized newsletters for registered and suscribed users which were sent twice per day via email. These developments were done almost end to end, because my tasks included gathering data from Google Analytics, real time measurement of page views, daily categorization of users depending on their news preferences based on semantic NLP analysis of read content, real time serving of personalized content with low delay and high RPM, most of these steps done using GCP utilities).In addition to my individual tasks I also performed as trainer and consultant for multiple teams in the company, helping to provide AI and machine learning solutions for diverse initiatives such as personalized job recommendations at elempleo.com, image analysis and labeling for the real state portal metrocuadrado.com, car valuation for the portal carroya.com.

Bogotá D.C. Area, Colombia

Colombia

At Universidad de Investigación de Desarrollo my two main roles were as lecturer and as researcher. As lecturer I led several courses from the computing sciences and electronics engineering programs, mainly on machine learning, signal processing, image processing, telecommunications, digital and analogic electronics, optimization and automation.As researcher I worked on automatic classification of human facial expressions using at that moment state of the art techniques (CNNs), automatic recognition of sign language, automation prototypes for the gas industry and other smaller projects.

Colombia

My research during my doctoral tenure was based on the automatic recognition of human facial expression. For that I evolved the static based algorithms POEM used for texture descriptions into temporal/volumetric features VPOEM (https://ieeexplore.ieee.org/document/6340576). The features obtained with VPOEM were treated as weak features in a metaclassification model. The results outperformed the at the time state of the art results using the standard Cohn-Kanade database. The full dissertation is available at https://noesis.uis.edu.co/items/f5e32563-2196-4e0c-8a30-466ca61227f9.

Bucaramanga, Santander, Colombia

Porto Alegre, Rio Grande Do Sul, Brazil

Lille, Hauts-De-France, France

Doctor Of Philosophy - Phd, Engineering

Master'S Degree, Signal, Images, Audio And Telecommunications

Master'S Degree, Electronics Engineering

Bachelor'S Degree, Electronics Engineering