CREG

Web platform for electrical operators in Colombia (PHP, Drupal)

Before finishing my career I started working as a research assistant in a High Performance Computing Lab. This was my first project for an external client: I Designed and developed a web application (Drupal, PHP) to evaluate strategic plans to reduce non-technical losses in electricity distribution for the Regulation Commission of Energy and Gas, CREG.

• Date: 2011
• Client: CREG
• Category: web development

Close Project

cvTraffic

Traffic sensor to count and classify vehicles using computer vision.

This was a project developed with the Colombian Administrative Department of Science, Technology and Innovation; COLCIENCIAS.

I co-developed (team of two) a traffic sensor to count and classify vehicles. We had all imaginable challenges coming from an outdoor and uncontrolled environment. The project was developed using early releases of OpenCV. I participated in idea generation to product deployment at traffic intersections. My teammate worked heavily in the hardware (processing unit and camera communication) while I was in charge of the software components, including the sensor communication. We both worked in computer vision for background subtraction and vehicle classification.

This is how our sensor was integrated with other components to analyze traffic congestion:

We integrated the sensor with traffic controllers and geographic information systems to visualize traffic state in different management systems across the city. Other small teams were responsible for different components, including the traffic controller, the communication system & the Geographic Information System.

We will later deploy a similar system for the entire city. For that project, instead of ArcGIS, we used a traffic simulation platform called TransModeler. I was responsible for its integration into the entire system.

Close Project

Traffic Central

Traffic sensor to count and classify vehicles using computer vision.

I designed the software architecture required for the Traffic Control Center in Pereira

I developed an application (.NET) using microkernel architecture to visualize the traffic lights of the city over a Traffic Simulation Platform (TransModeler), enabling traffic simulation using real time information.

Co-developed an application to manage the timing of traffic controllers for the city of Pereira following NTCIP communication protocols.

These products allowed us to be part of the Pereira Innovation and Technological Development Center (Colombia), creating an university spin-off focusing on Intelligent Transportation Systems.

Close Project

smaps

Web platform for Intelligent Transportation Projects

I led the implementation of 'Regional Intelligent Transportation Systems Architecture for West Central Metropolitan Area' with a web platform (Node.js, OpenStreetMap, Common Alerting Protocol) integrating regional transport information from multiple data sources including traffic data, road emergencies & alerts, traffic light status, security cameras, public transport among others.

Smaps is currently under development by a different team. You can check their website here.

• Date: Dec. 2015
• Client: CIDT
• Category: Backend/web development

Close Project

Bariatrack

Software engineer for mobile app.

We developed a prototype required to evaluate the evolution of bariatric surgery' patients. I had to find the way to adapt the BAROS (Bariatric Analysis and Reporting Outcome System) for Colombian users in a digital environment. I led the development team and I was in charge of direct interaction with the client.

• Date: Dec. 2016
• Client: Mednovation (Medellín, Antioquia)
• Category: Mobile

Close Project

Colombian Armed Conflict

Interact with Colombian Armed Conflict data from Centro Memoria.

At this point I knew I wanted to pursue a data-driven career. I started to study visual analytics not only because it was a missing skill in our team. Besides taking responsibility for visual components in our projects at work, I continued learning in my free time developing side projects.

This is a data visualization project about warlike actions in the context of the Colombian armed conflict. In 2016 I sent this project to Centro Nacional de Memoria Histórica, asking for more data and proposing new ideas for visualization. They told me they were developing a project covering the topic. It was later published at 'Rutas del Conflicto', which I think lacks data-humanism. I agree with Lisa Charlotte Rost when she says we need to create visualizations and make people feel empathy about these topics.

• Date: Nov. 2016
• Client: Side project
• Category: information visualization

Close Project

Geo Tweets

Tweets visualization according to topic and geolocation.

This is another side project. Tweets visualization developed at Bios to practice and showcase some technologies. You can search up to four (4) different topics to see places where people are tweeting them the most. You can also filter topics by sentiment.

I designed the interface and lead the team for project implementation. I developed d3.js and leaflet related components.

All my side-projects in visual analytics would later be crucial to obtain a scholarship from the European Union for the Master in Big Data Management & Analytics, which I started in 2018 :)

• Date: 2017
• Client: Internal project at BIOS
• Category: information visualization

Close Project

UDAPTOR

Big Data startup business plan & prototype

In our Master we developed a data portability tool, we called it UDAPTOR. The project takes advantage of the General Data Protection Regulation (GDPR) to transfer data between service providers. Our idea was to facilitate data portability converting collections of files from one provider (e.g. Spotify) to another (Deezer, Apple Music, etc.). I participated from idea conception to the prototype implementation, responsible for the Knowledge Graph mappings required to transform the source files to the target files. I also developed the PoC using Spotify, improving music recommendations for users coming from Apple Music.

Later, two members of the original team continued the project with their BDMA Master Thesis. Here the official website, where you can find more info.

• Date: July. 2019
• Location: Barcelona, Spain
• Client: UPC - Class Project
• Category: data engineering, data integration, semantics

Close Project

Twitter sentiment analysis

Sentiment analysis & Visual Analytics.

With the objective of getting information about the Perception of migrants on social media, we gathered Twitter data during the months of August, September and October 2019. We looked for tweets in English, that were not re-tweets and that contained one of the following words: migrant, immigrant, refugee and asylum seeker.

Visual Analytics Project. Course by Petra Isenberg at CentraleSupelec.

I designed the initial sketch for storytelling and implemented 80% of the visuals using d3.js.

The project is available at this Github site where you can find a data analysis report.

• Date: January. 2020
• Location: Paris, France
• Client: CentraleSupélec - Class Project
• Category: information visualization, data science, visual analytics

Close Project

GeoBERT

Language Model for GeoSciences.

For my BDMA Master Thesis & Data Science Internship at Schlumberger Montpellier.

I built a Named Entity Recognition (NER) System for the Oil&Gas Industry.

First, I developed a distributed NLP pipeline using Apache Spark NLP for weak data labeling with distant resources (dictionaries, RegExp, ontologies) making the pipeline extensible to new named-entities and suitable for Big Data scenarios in other domains.

Using the noisy labels, I trained Deep Neural Network language models (BERT, DistilBERT & DistilRoBERTa) with a two-step fine-tuning process, improving accuracy in hard-to-learn named entities and showing promising results to solve the polysemy problem in our domain.

The project was developed in a cloud environment (Kubernetes, GCP) & it was managed using developer tools for Machine Learning (nbdev, Weights & Biases). The team continued the project at Schlumberger to incorporate more entities, aiming to build a complete Language Model for GeoSciences (GeoBERT)

At the beginning, this project was especially challenging since it was developed during the COVID-19 pandemic, but I was able to improve my capacity to keep focused and motivated

You can check the project presentation over here.

• Date: September. 2020
• Location: Montpellier, France
• Client: Schlumberger, Master Thesis
• Categories: Natural Language Processing, data science, Apache Spark, spaCy

Close Project