Projects
Animal Breed Classifier
A Streamlit application that allows users to upload images of dogs, cats, or birds and see what specific breed/species the animal is. Supports 120 dog breeds, 31 cat breeds, and 59 bird species.Code
LinguifAI
A desktop application allowing users to train classification models on their data or use pre-trained classifiers to gain insights about their data. Users may also talk to a chatbot to ask questions about their data. No coding experience required!Code
Optimal Course Scheduler
Developed an algorithm to optimize my course schedule, taking into account my preferred courses, their seasonal availability, and the specific requirements for my two majors. This algorithm efficiently identified the quickest path to completing all academic requirements, which helped make my educational journey faster and allowed me to have room for extra classes. While the system is currently customized for my situation at Cal Poly, it can be adapted for other students at the same institution. However, because it relies on scraping the university's course catalog, its applicability to students from other universities is limited.Code
Microstructure Reconstruction and Generation
Designed a GAN to generate new microstructure images given a low number of high-resolution samples. The GAN is also able to accept class labels, which in this case were based on the percent strain applied to the material. Some important features of the GAN include differentiable augmentation, a custom loss function combining LPIPS with microstructure-specific metrics, and skip-layer excitation blocks.Code
Crack Segmenter and Measurer
A Streamlit application that allows users to find the exact length of a crack simply by inputting an image. The application allows the user to calibrate the camera (find the pixel-to-inch ratio) and then uses an algorithm to calculate the full length of the crack in inches. This project was completed based on a request from an infrastructure inspector looking to automate the process of finding cracks in buildings, sidewalks, and other pieces of infrastructure.Code
SARFormer (AI for Search and Rescue)
Developed a multi-modal transformer model, trained on real missing persons cases, using a two-stage process: pre-training with a masked modeling objective followed by fine-tuning to predict the location of missing persons. The model integrates satellite imagery, text descriptions, and tabular data (weather and time) as inputs to generate a probability map indicating the most likely location of a missing person.Code (coming soon)
Analysis of Technological Anglicisms in Different Spanish-Speaking Countries
Built a site that displays results of research regarding the use of technological anglicisms in Mexico, Spain, and Argentina. Also includes a statistical analysis I performed on news sites from those three countries using Python.Site
Animal Breed Classifier
A Streamlit application that allows users to upload images of dogs, cats, or birds and see what specific breed/species the animal is. Supports 120 dog breeds, 31 cat breeds, and 59 bird species.Code
LinguifAI
A desktop application allowing users to train classification models on their data or use pre-trained classifiers to gain insights about their data. Users may also talk to a chatbot to ask questions about their data. No coding experience required!Code
Optimal Course Scheduler
Developed an algorithm to optimize my course schedule, taking into account my preferred courses, their seasonal availability, and the specific requirements for my two majors. This algorithm efficiently identified the quickest path to completing all academic requirements, which helped make my educational journey faster and allowed me to have room for extra classes. While the system is currently customized for my situation at Cal Poly, it can be adapted for other students at the same institution. However, because it relies on scraping the university's course catalog, its applicability to students from other universities is limited.Code
Microstructure Reconstruction and Generation
Designed a GAN to generate new microstructure images given a low number of high-resolution samples. The GAN is also able to accept class labels, which in this case were based on the percent strain applied to the material. Some important features of the GAN include differentiable augmentation, a custom loss function combining LPIPS with microstructure-specific metrics, and skip-layer excitation blocks.Code
Crack Segmenter and Measurer
A Streamlit application that allows users to find the exact length of a crack simply by inputting an image. The application allows the user to calibrate the camera (find the pixel-to-inch ratio) and then uses an algorithm to calculate the full length of the crack in inches. This project was completed based on a request from an infrastructure inspector looking to automate the process of finding cracks in buildings, sidewalks, and other pieces of infrastructure.Code
SARFormer (AI for Search and Rescue)
Developed a multi-modal transformer model, trained on real missing persons cases, using a two-stage process: pre-training with a masked modeling objective followed by fine-tuning to predict the location of missing persons. The model integrates satellite imagery, text descriptions, and tabular data (weather and time) as inputs to generate a probability map indicating the most likely location of a missing person.Code (coming soon)
Analysis of Technological Anglicisms in Different Spanish-Speaking Countries
Built a site that displays results of research regarding the use of technological anglicisms in Mexico, Spain, and Argentina. Also includes a statistical analysis I performed on news sites from those three countries using Python.Site