During my university internship, I developed a mobile app using Expo/React Native and a REST API with PHP for Ito-e, a company specializing in providing services to construction companies for efficient project management and tracking. The application empowers Ito-e users to oversee project progress and conduct on-site inspections, allowing for organization by projects, typologies, batches, and more. Users, with different roles and permissions, can perform inspections in both online and offline modes. They can capture notes, photos, GPS location data, and other relevant information and can accept or reject completion indications provided by other users. The development posed a significant challenge due to the vast and intricate database and resource-intensive operations required for measuring project progress and implementing complex search filters. To address this challenge, I devised a solution for offline functionality, implementing a method to compress and simplify project information and images, storing them in a local database using SQLite. Additionally, I optimized and replicated the sophisticated server-side searches on mobile devices with limited processing power. Once an internet connection was restored, all locally captured data seamlessly synchronized with the server. The company managers were impressed with the outcome, considering my work at a senior level. They mentioned that due to the quality of the product, they felt confident in expanding their services to international clients.

I developed a 3D augmented reality mobile game using Unity for Unicorn Games, a company that combines tabletop games for children with technology. In this game, players engage in shooting at dynamic 3D targets that seemingly hover on the screen, seamlessly integrated with real-time images captured by the camera. This creates the illusion that the targets exist within the same room. The aiming mechanism involves utilizing the phone oscilloscope. Simultaneously, I contributed to enhancing the visual effects of another augmented reality game within the company. This specific game, centered around Lego elements, manifested on the screen when kids directed the camera towards a game card.

Google Drive REST API - 2019

I created a concise website and REST API using Node.js for an Argentinian company named Red Tecnológica. The application offered an interactive platform for students to explore and interact with Google Drive functionality. This included features such as adding, removing, and modifying files and directories, as well as sharing files and other related functionalities.

Body Gestures Controller - 2020

I created a Linux desktop application for one of my freelance clients. This application enables users to map body gestures captured by the camera to system events, allowing functionalities such as controlling the pointer or managing racing games using hand gestures. To achieve this, I used the PoseNET model, enabling precise tracking of body part positions captured by the camera, meeting the client's desired level of precision.

I developed a web application for a small business dedicated to providing tutoring services to pre-university students from their homes. The website comprised three panels: one for students, allowing them to view a calendar with available tutor hours for specific subjects, schedule appointments, and make payments through Transbank Webpay or in cash; another panel for tutors, enabling them to add the subjects they can teach and their schedules; and a third panel for administrators, providing them with the ability to analyze student and tutor profiles/reviews and manage finances. Additionally, the website calculated the transportation cost for tutors to reach the students' homes.

File Processing Chain - 2021

I developed a desktop application using Qt and implemented a Python microservice for a freelance client. The project involved creating an application that allows users to select two directories—one to monitor for incoming files (from another local application) and another to receive the processed files from an external service. Essentially, the local files were sent to a Python microservice, which added them to a RabbitMQ queue. Subsequently, the files were processed by an external service and sent back to the microservice as processed. The desktop application then fetched the microservice and stored the processed files in the second directory. Additionally, the desktop app displayed a table containing file information and the processing status.

The goal of this project was to create a convolutional neural network for accurately classifying hand images featuring various gestures. Subsequently, the performance of this network was compared with a fine-tuned ResNet18, specifically retrained on its fully connected layer. Following extensive testing and training using diverse architectures and image preprocessing techniques, I successfully developed a model that demonstrated notably higher accuracy than ResNet18, particularly in the context of hand gestures recognition.

This project involved the development of a video encoder and decoder, along with adjustable noise elimination filters, to allow video transmission across both high and severely constrained bandwidths. Within this project, various techniques were applied, including the use of the 2D FFT, FIR, or IIR filters to eliminate sinusoidal noise, methods to address saturated pixels, and diverse compression techniques such as lossy compression using the cosine transform, perspective transformation analysis using the mean squared error, scaling or translation between frames, and redundancy reduction using Shannon entropy theory.

The objective of this project was to determine if it is possible to predict the number of billionaires per country based on variables such as population size, GATT, and GDP by modeling the problem with a Poisson regression. The project involved finding the model weights using various optimization methods and analyzing the results using resampling techniques such as bootstrap.

In this project, I was tasked with enhancing the TFTP 1350 file transfer protocol to include support for user authentication, directory navigation, and simultaneous packet and file transfers. The protocol specifications were documented in the form of an RFC, and I additionally provided an example implementation using Python.

This project focused on analyzing, identifying, and filtering the noise present in signals generated by electrograms. To achieve this, an analysis was conducted in the frequency spectrum using the Fast Fourier Transform (FFT), spectrograms, and various types of filters were then applied. Subsequently, the effectiveness of Finite Impulse Response (FIR) and Infinite Impulse Response (IIR) filters was compared.