1st place for Best Real World Application - Hack@Brown 2019

  MindSpaces is a responsive guided meditation experience in virtual reality. An OpenBCI EEG brain sensor worn by the user transmits data in real-time to update the VR environment and personalize the meditation guidance. When a user's brain is highly active and stressed, their world consists of gently pulsating geometric shapes. They might receive tips such as "Come back to your breath." As a user calms down, the shapes shrink into tiny dots of light, like a night sky. The guided meditation is translated via text that floats in front of the user.
  The goal of this project is to give users new tools to engage with meditation and improve their mental health. By showing users a visual representation of their current mental state, users benefit from a more clear self-awareness about how they are feeling. This also provides users with a more tangible and engaging way to experience how their mind changes over the course of a meditation. Moreover, by being immersed within a beautiful and calming virtual environment, and by receiving meditation guidance based on their moment-to-moment needs, users are able to develop a toolkit for actually engaging with and managing their different emotional states. Lastly, Mindspaces allows for certain kinds of progress in meditation to be more easily tracked over time, providing tangible incentives and feedback for day-to-day meditators frustrated with ambiguity about if meditation is helping them.
  I personally formed the original concept for the project, implemented the server to modulate VR parameters based on live-streaming EEG data (Python Flask), designed the guided meditation, and oversaw the single component analysis of EEG data (Python) and all VR design (A-Frame).

  To learn more about Mindspaces, check out our DevPost!

  Maps is web app replica of popular navigation apps, such as Google Maps, which I built full-stack using Java and JavaScript. The program provides efficient directions, accounting for real-time traffic, on an interactive map. My implementation uses real-time traffic data to find the fastest routes at any moment. For finding routes, I implemented a K-D Tree for nearest neighbor search, A-Star shortest path search, and street and traffic SQL databases with optimized caching. The program runs using a Spark server and has a complex GUI built using HTML Canvas, FreeMarker, and JavaScript.

  Autocorrect is a software and AI project that generates suggestions for words that a user may have meant to write in real-time as they type, similar to what Apple does for text messaging or what Google does for Search. My version is written in Java and accomplishes two primary backend tasks: generating suggested words and then ranking those suggestions from most likely to have been meant by the user to least likely.
  For suggestion generation, I implemented a Trie data structure used for prefix matching (if "compu" is typed, "computer" might be suggested) and for whitespace checking (if "sandcastle" is typed, "sand castle" might be suggested). I also implemented Levenshtein Distance Editing to measure how many edits a word needs to become another word. For instance, if a user types "philosphy", my program recognizes that this is only one letter insertion away from "philosophy", so "philosophy" will be a suggestion.
  To rank suggestions, my version first considers keyboard distance. For instance, "vomputer" is only one character swap away from "computer", and 'c' and 'v' are right next to each other on the keyboard, so "computer" will be a very likely suggestion. My program also considers how frequently words are used (unigram values) and how frequently they are used after other words (bigram values) in several novels that I used as data.

"For every 100 pitches that I see, this one goes right to the top. Somebody will invest in this, I assure you." - Bob Stearns, Managing Director at Sternhill Partners Venture Capital

  Cherry is an entrepreneurial venture developed by me and three other students in Brown's reputed Entrepreneurial Process course. While we ultimately chose not to pursue funding, we accomplished all of the steps necessary to bring Cherry to the point of being a worthy candidate for seed investment, culminating in pitching live to VCs in April 2020 and being encouraged that we were ready for funding. You can find our pitch deck linked here.
  Cherry is both a consumer-facing app and a POS integration. Cherry’s app gives restaurant and bar-goers access to their checks right from their phones. Users can open, split, and pay for tabs with ease and at their own convenience. Cherry allows users to leave restaurants and bars on their own time and removes the kinks involved with splitting the check amongst party members. For businesses, Cherry seamlessly integrates with existing POS systems to increase efficiency and revenue. Cherry increases server efficiency, customer satisfaction, and average tab and tip values without the need for any additional hardware (iPads, new POS system) or operational complications.
  I personally formed the idea for Cherry, led our product development process (features, design, financial considerations, technical research; we did not create the product), and shared responsibility in designing our business model, marketing plan, market research, pitch development, and development of our 30-page business plan.

Facebook And Amazon Antitrust Research Project

  As my final project for my Cybersecurity Ethics course taught by Professor Deborah Hurley, I worked with a small group on a research project seeking to examine and develop a proposal regarding Big Tech's potential antitrust violations and how regulation might be enforced. We narrowed our efforts to focus specifically on Facebook and Amazon. As part of our examination, we researched different economic schools of thought on antitrust behavior, the history of antitrust regulation in the U.S., Facebook and Amazon's history of potentially anti-competitive behavior, the current political and legislative climate in the U.S., and different attempts at antitrust around the world, particularly in the European Union. We concluded with a synthesis of our research and proposal for how the U.S. might move forward. We presented our work to the class and received a grade of 100% for our work.
  This project, as well as this course overall, gave me an appreciation for the complexity of the current landscape at the intersection of technology, traditional human systems (economies, governments, legal systems, etc.), and human beings in general. I left this course both with a deep concern for the future, but also inspiration because of the intelligent and dogged pursuits of others laying the path toward humane, wise, and democratized solutions for the use of technology.

Original Large-Scale Web App

  Payfi is a large-scale web app that I and two teammates designed and engineered full-stack. Engineering PayFi involved a notable amount of independent research and creativity, including testing out different OCR computer vision technologies, creating an original Venmo API by copying Venmo's own post request structure, and learning how to use and implement a MongoDB database.
  Payfi addresses the problem of splitting bills with friends. Payfi allows users to take a photo of a receipt, automatically produce an itemized list of receipt items, assign and split items between friends, automatically calculate all who-owes-who's, send multiple Venmo payment requests in one-click, and track the completion of Venmo transactions.
  I personally formed the idea for Payfi, led the design process (features, screen layout, etc.), and was responsible for implementing our databases, backend data management, and general backend structure. I also helped problem solve OCR and Venmo API issues.
  Our team received a 100% for our work, being judged on engineering difficulty, engineering design choices, entrepreneurial practicality, degree of independent research needed, aesthetic design, and creativity. Though our team did not continue to pursue this project, PayFi was both inspiration and partly a technical proof-of-concept for my later project, Cherry.

  As assignments in my Deep Learning course, I implemented all of the following types of deep neural networks using both TensorFlow and PyTorch: MNIST digit classifier (using only NumPy), Convolutional neural networks to classify images of cats vs. dogs, Recurrent neural networks (RNN) with and without Long Short-Term Memory (LSTM) to generate coherent text, Transformers for language translation, Message-Passing neural networks (MPNN) to recognize graph structure of molecules to predict their activity against cancer, Reinforce with critic algorithms to learn the computer game cartpole, and General Adversarial Networks (DCGAN) to generate artificial images of faces.
  All of my networks performed with high accuracy considering the computational resources available. I finished the course with an A. I left this course with a great appreciation for the ingenuity underlying deep learning, as well as with a genuine interest in and concern for the impact that these technologies will have on humanity.

  As assignments in my Artificial Intelligence course, I implemented all of the following in Python: Decision Trees, SARSA Reinforcement Learning, K-Nearest-Neighbors clustering, Hidden Markov Models (for noisy touchscreen data), and Linear Regression Learning. I finished the semester with an A in the course based upon the optimality and accuracy of my implementations.
  For my final project, I developed an artificial intelligence bot designed to play the renowned computer game Tron. The bot, implemented in python, uses alpha-beta pruning with a minimax strategy, has a cutoff with a voronoi heuristic, and uses a depth first search for end-game space-filling strategy. This bot has a 90% victory rate against human opponents.