A full-stack web application that eliminates the tedious process of manually tracking college grades. Users simply paste their course syllabus, and the app instantly generates a fully customized, interactive grade calculator.
Behind the scenes, the application leverages the Gemini LLM API to process unstructured syllabus text. Using carefully engineered prompts, it identifies grading categories, weights, and assignments, converting the raw text into highly structured JSON data to dynamically build the frontend calculator.
To make the tool functional for a full semester, the app utilizes Supabase (PostgreSQL) as a Backend-as-a-Service (BaaS). It handles secure user authentication and persistent state management, allowing students to log in, save multiple course calculators to their profile, and update their grades via full CRUD operations as the semester progresses.

/* Github Repo */

/* Live Link */

A 2-in-1 tool that provides unique insight into Formula One races.
The radio explainer transcribes radio messages and provides an AI explanation for users just getting into the sport, considering key context in its response. The model was fine-tuned using over 600 radio messages from the 2023 season.
The lap visualizer allows the user to select any two drivers' laps to compare. The drivers' realtime telemetry and racing line are then played out, showing users the fine margins between great and legendary drivers. More technical details are available on the website.

/* Github Repo */

/* Live Link */

A C++/SFML-based Blackjack simulator that trains players in decision-making and card-counting techniques. After each move (hit, stand, double, etc.), the app evaluates whether the choice was statistically optimal and provides instant feedback.
Users can also enable true count tracking and live player edge (for real-time card-counting practice), which is calculated by running 100,000 Monte Carlo simulations per hand, using optimal strategy on the remaining deck composition.

/* Github Repo */

A full-stack website for users of card-collecting apps Pacybits/Madfut to post and search for trade requests, helping users complete their collection more effectively than ever. Started in 2022, revamped in 2024.
Uses a Node.js backend connected to a MongoDB database to store user posts. A custom web scraper was used to generate an array of over 6300 objects and download images for each. Vanilla HTML, CSS, and JavaScript make up the frontend. Deployed on Heroku via GitHub.

/* Live Link */