My Projects + TODOs
I'm going to include my startup projects here too, rather than in the experiences page. I don't sleep as much as I should, and this is why. A lot more AI projects on the bottom, if that's your jam. If you like one of my TODOs, let me know! Or even better, build it and show me. I'm pretty open source with ideas because at some point you realize the amount of execution arbitrage it really takes for something to be great.
TODOs
AI Interaction Tracker
Let's be clear. AI is the future. AI is the future due to accessibility of infromation. NOT because you can automate everything with an AI agent and replace all workers. God, I see so much doom-fantasy online about how AI will take our jobs. It won't. It's statistical autocomplete. With that being said, AI will greatly transform education. Most students use it now anyways to learn. Why not govern and build a tool that can analyze someones conversation with AI to see if they're using it for their benefit by delaying gratification and using it to help them digest concepts + apply learning or simply just trying to be lazy and have a model do all their work. A tool like this can rapidly transform the education system, allowing incredible learning and innovation. It can also be used by companies for hiring, such as leetcode style interviews, or analysis of coding agent use in junior devs, etc.
Projects
Agosec
Built an AI-first iOS keyboard extension + container app with Agent Mode chat, screenshot context import, and StoreKit2 subscription gating. The keyboard sends OpenAI-format `/v1/chat/completions` requests to my LiteLLM backend fork, using custom implemented auth. It bootstraps proxy keys for model routing with default Deepseek-chat but also BYOK for OpenAI/Anthropic/Gemini models.
Monte Carlo Right-Turn RL Agent
Built a custom racetrack reinforcement learning environment in pygame where a Monte Carlo agent learns to navigate randomized R-shaped tracks. Implemented episode-based policy evaluation, epsilon exploration, real-time training visualization, and interactive controls for generating tracks, pausing, and resetting runs.
CS138 RL Ticket Pricing Agent
Built a dynamic ticket pricing agent for live events (Tufts COMP 138 final project) using both Tabular Q-Learning and DQN. Trained against a data-driven ticket demand simulator inspired by SeatData.io data, compared against heuristic baselines, and added visualization and checkpointed training workflows for long-run experiments.
LinkedIn Student Email Scraper
Built an automated LinkedIn student profile scraper that logs in, applies school filters, extracts profile metadata, and stores results in SQLite with logging and rate limiting. Extended it with an outreach pipeline that generates likely school email patterns and sends tracked Gmail SMTP campaigns with delivery status updates.
Orange Theory AI Agent
Used custom MCP server built around unofficial orange theory client library to build an AI agent for my gf that acts as her personal Orange Theory assistant. Uses A2A server system as well, with Next.Js frontend demo.
LLM from scratch
I built a GPT style transformer LLM from scratch. I used the PyTorch framework to build the model. Also used Wizard of Oz, and Vocab data. Really learned a lot about the math behind LLMs, and how they work. Also learned why I really need Nvidia GPUs. Google Colab saved my butt.
Options Pricing API
I was inspired to make this after my SIG discovery day event Sophomore year, where I learned about the importance of RPC latency in the quantitative finance industry. I've always been interested in options pricing, growing up playing with options volatility, and love the math behind different pricing models. Hudson River Trading has a lot of articles of things you can do with gRPC servers. For this project, I built a gRPC server that can price options using multiple pricing models. I used the Black-Scholes model, the Heston model, and the Variance Gamma model. I'm not really sold on Black-Scholes, as volatility tends to be dynamic and random.
HelperU
Co-built a platform for adults to hire college students for gig work, such as tutoring, dog walking, and more. Imagine TaskRabbit, but for college students. Sold to JouwAIPartner in December 2025 to be turned into an AI Skills Marketplace.
Tufts Free Food Module
I've built a module for Tufts University's mobile app that allows students to find free food on campus. I worked with the Tufts Unviersity Friedman School of Nutrition Science and Policy to build this, and we were able to pitch and partner with Tufts Technology Services to get the module built. It's now in the Tufts mobile app!
TouredIt
For a very long time, this was my child. TouredIt is a startup focused on providing personalized tours of college campuses. The idea is that if your child is interested in a certain field, hobby, or social group, they should be able to tour college campuses with students exactly like them. I built the website, the backend, the tour guide portal, and everything in between. CPO, CTO, whatever you call it, imagine putting out technical fires 24/7. Shoutout Joshua Bernstein, my partner, for being a great example and friend. He is one of the most talented and resourceful people I know.
Nephwz
Not a technical project, but something I'm extremely proud of. With a friend, we built a social media page named Nephwz. We posted curated art, music, and culture content. From Winter 2023 to Fall 2024, we amassed a little over 22,000 followers and got a 55M view reel. Unfortunately, with being a full time student athlete, recruiting for internships (Leetcode please save me), running TouredIt (We had marketing interns!), and launching the Tufts Free Food Module, I could not keep up with the content and had to stop. Many of the people we started along with have now become so massive. As much as I hate to admit it, watching our competitor, Flyhysteria, grow from 1000 to 78k followers was a fun experience. I'm not sure if I'll ever be able to top consitently getting 3000 likes on a post.
Sports Twitter Sentiment Analysis
My first ever full stack technical project, summer of my Freshman year. I built a sentiment analysis tool that scrapes Twitter data about the Philadelphia Union, analyzes the sentiment of the tweets, and displays the results in a dashboard. It parallizes it with recent news articles and match results about the Philadelphia Union to give a holistic view of the team's sentiment. I also tracked the rate of change in twitter sentiment, with cron jobs to text my cellphone if there was a significant change. The idea is that Sports Traders could have an advantage on betting lines before the market moves the line. Unfortunately, this was around when Twitter turned to X, and nuked their developers portal + any scraping ability, so this project became deprecated.
AI Food Similarity Search
AI Powered tool to find nearby restaurants based on a picture of uploaded food. Ranks and displays restaurants by the restaurant's food image similarity. Backend built with Python and Flask framework. Hosted on AWS using Ubuntu EC2 Instance. Uses pre-trained Salesforce Blip model for image recognition and Vgg16 model for cosine similarity analysis. Uses bs4 for restaurant photo scraping and Scraping Bee API to bypass IP blocking. Yelp, I'm sorry for the love hate relationship we have. I think my computer IP was blocked from Yelp's website for scraping over 1000 times. I was just a broke college student in my dorm room, the project is deprecated anyways.
AI Notes App
AI Powered notes app to keep track of all your notes while using AI to optimize key insights within them. NodeJs and ExpressJs backend w/ PassportJs for google authentication. MongoDB for Note and User Data, deployed with Heroku, and GPT-3.5 for note optimization
Bunzo Chat
Ok. Did I build an custom AI RAG Chatbot for a Japanese Literature class, so I could spend less time reading the book and more time writing the paper (I did not plagarize, I used it for citing sources). Yes. I did. It's called Bunzo Chat, because the main character of 'The Drifting Cloud' is named Bunzo.
Flowpa.co
Freelance two-person side business building websites, AI chatbots, and automated marketing for small businesses. Partnered with Philmont FC soccer club to implement an AI chatbot solution (using OpenAI and FAISS vector database). Partnered with Forward Progress Injury care to run google ad marketing. Designed and built PHP oriented website for nephwz.com