My Projects

Generating cover letters with the GPT-2 model and deployed on Google Cloud

Everyone knows that writing letters of recommendation is hard work and not everyone's cup of tea, but on the other hand it is essential to be able to apply for some jobs. Why not use artificial intelligence to do so?

I have created an application using the GPT-2 model trained to generate letters of recommendation. The application is built using flask and docker, the service is deployed on Cloud Run (GCP).

How people talk about Coronavirus using Kafka and Twitter

This website was developed using Twitter API, Google Cloud and Apache Kafka to view the messages related to Coronavirus in real-time. I started this project since I thought that it would be interesting to analyze how people are talking about the Coronavirus Crisis on Twitter.

The producer tracks, gets and sends tweets about Coronavirus to the Kafka Cluster that I deployed on Google Cloud. The consumer pulls and processes the messages from Kafka. At the end, the front-end represents this data in a map. The project was deployed on Google Cloud and it was programmed using Python and Javascript.

This is one of my last side projects and I am constantly implementing new features. I would like to implement a sentimental analysis and a named entity recognition to extract relevant tokens to analyze these messages. This project can be adapted to any topic that you would like to track from Twitter.

App for iOS (Swift) to colorizer images in B&W

App for iOS developed using Swift to colorize images that were originally in black and white. The idea behind this project was, in addition to learning swift, to create a mirror to the past, allowing us to publish and share our old b&w images.

The app uses an open-source model to colorize the images. Although this model was only enabled to run on GPU, I managed to make predictions on CPU with the same performance (I am contributing to this open-source project). This allows me to deploy this model in a more economic instance on any cloud. I hope to commercialize my app within the next few months.

Cloning human driving behavior using a DNN

I used a simple car simulator in order to clone the human driving behaviour. During the training phase, I navigated the car inside the simulator using the keyboard. Simultaneously, the simulator recorded training images and respective steering angles. Then I used those recorded data to train a neural network.

Named Entity Recognition in travel-related documents using RNN

Project to create a named entity recognition (NER) for travel-related sentences. I developed this project inspired by how Gmail extracts information from the travel-related emails and automatically inserts the event in your calendar. It developed it using Keras and the model is deployed as REST API.

Call Center to analyze automatically if a person can be infected with Coronavirus

I have managed to create a Call Center using Amazon Connect and Amazon Lex Bot. The bot asks to the customer several questions to analyze if this person can be infected with Coronavirus. This is just for study purposes, it is NOT official

You can try it here: +61 2 8311 7630

Service for generating text with the GPT-2 model and deployed on Google Cloud

GPT-2 is a large transformer-based language model with 1.5 billion parameters, trained on a dataset of 8 million web pages. GPT-2 is trained with a simple objective: predict the next word.

The main objective of this project is to create a service for generating text using the GPT-2 model. We've built the application using flask and docker, the service is deployed on Cloud Run (GCP).

How to create a Neural Network from Scratch - No frameworks

This is a simple recreation of this neural network playground, a tool that provides a better understanding of Neural Networks. In this project, I built a neural network from scratch, without using any framework (Keras, Tensorflow, ...) just our own functions. It is very useful for understanding the concepts and basic parameters of the neural network.

Alexa Skill using Lambda Functions and Web Scraping

I developed this Alexa Skill to understand how Alexa works and gain a much deeper insight into Lambda Functions. It makes use of web scraping to extract proverbs from one website. It was built using the Alexa Skill Kit SDK for Python and it is deployed on AWS.

Measuring social distance with TensorFlow

Today, unfortunately, everyone is familiar with the term "social distance". It’s something we will have to live with for a while until everything returns to normal. I have tried to develop an application using the TensorFlow Object Detection API for identifying and measuring the social distance between pedestrians.

Implementation of a Deep Q-Network (DQN) model

This project uses a Unity environment to create a map with bananas scattered around. There are 2 types of bananas: yellow bananas and blue bananas. The objective is to train a Deep Q-Network (DQN) to collect yellow bananas and avoid blue bananas.

Convolutional Autoencoders to remove the noise in documents

Challenge from Kaggle to remove the noise of images of scanned text that has seen better days. I have used a Convolutional Autoencoders to improve the readability of these documents.

Implementation of "Image Style Transfer Using Convolutional Neural Networks" paper

I have recreated the style transfer method of this project: Image Style Transfer Using Convolutional Neural Networks. The features found in the VGG19 Network are used to transfer the style between pictures. Project developed using PyTorch as well as Keras.

I have used my dog, called Roscón, as a model for this experiment! He was rewarded with a treat.

Brief approach using TF Object Detection API for finding Wally

I remember when I was a child I spent hours trying to figure out where Wally was. I had a lot of fun with these books but sometimes I lost my patience and I anxiously awaited the development of a system that could find Wally automatically.

Currently, thanks to Artificial Intelligence this is possible, so this project entails a quite easy approach to the application of the Tensorflow Object Detection API to find Wally.

Noise for improving Deep Learning models

Acquiring more data is a very expensive and arduous task. However, we can apply some techniques (regularization methods) to ensure a better performance of our model.

In this project, I focus on the use of noise as a regularizing method in a neural network. This technique not only reduces overfitting, but it can also lead to faster optimization of our model and better overall performance.