Projects

Built with React (web) and React Native (mobile), powered by a FastAPI backend and Supabase database. The app features specialized progress tracking for bodyweight exercises through challenge and assistance metrics, such as resistance band levels and equipment height adjustments, solving the unique challenge of measuring calisthenics progression beyond traditional weight tracking and rep counting.

Built with Databricks, PySpark, and Delta Lake, connected to a React dashboard through a lightweight API layer. Raw FHIR Bundles from Synthea are parsed and flattened using PySpark into four Delta tables (Patient, Encounter, Condition, Observation), then exposed through a Databricks SQL Warehouse. A small API wraps that endpoint to serve clean, purpose-built JSON, which a React dashboard consumes to render a single, focused visualization. The project started as a focused Databricks/PySpark learning exercise and was extended into a small end-to-end application, prioritizing depth and correctness in one working slice over a broad analytics suite.

Built with AWS S3, Glue, and Redshift Serverless, connected through a private VPC with least-privilege IAM roles scoped per service. A CSV uploaded to S3 is cataloged by a Glue Crawler, transformed by a Glue ETL job, and loaded into Redshift via COPY, with all service-to-service traffic routed over a private S3 Gateway endpoint rather than the public internet. The project intentionally skips a full medallion architecture and real business logic, focusing instead on proving every piece of infrastructure—networking, permissions, service connectivity—works correctly end to end. It's the foundation for a follow-up project that expands the same pipeline to ingest and process real healthcare data in FHIR format.

This is a simple data engineering ETL pipeline project using the E-Commerce Purchase History Dataset from Kaggle. The raw CSV is extracted, transformed, normalized, and loaded into a PostgreSQL relational database. This pipeline establishes a foundation for downstream data analysis and decision-making.

Built with Python and OpenAI's GPT-4o-mini using structured output, with Pydantic for data validation. The tool extracts recipe text from images—such as screenshots or photos of handwritten or printed recipes—parsing ingredients, quantities, prep and cook times, and calorie estimates into a clean, structured markdown document. Solves the tedium of manually transcribing recipes from photos or screenshots into a consistent, readable format.

Built as a learning project to explore LLM integration in a production-like architecture — and as a reference implementation for an AI workout generation feature planned for Torque, my calisthenics workout tracking app. This project is currently in active development.

After creating a username, users can instantly video chat with other online participants. The platform uses Socket.IO for signaling and connection management, while establishing direct RTCPeerConnection instances for media streaming between clients. Redis is used for in-memory session and connection state management. The project was created as a deep dive into real-time communication protocols and the challenges of implementing WebRTC. Currently updating tests, moving the frontend from Create React App to Vite, and improving code organization and error handling.

The project replaces the original Create React App setup with more contemporary tooling and dependencies, while elevating the visual design. Developed as a practical exercise in leveraging AI tools for efficient development—streamlining the modernization process without sacrificing code readability and maintainability. Demonstrates thoughtful integration of AI assistance while maintaining clean, professional code standards. This project is currently in active development.

The application delivers real-time feedback distinguishing between correctly positioned digits (bulls) and correct digits in incorrect positions (cows). Developed using test-driven development practices to strengthen core programming fundamentals and ensure robust, maintainable code without framework dependencies.

Built with Django and Beautiful Soup, the application parses HTML content to identify and extract structured recipe data, presenting it in a clean, readable format. Developed to master web scraping techniques with Beautiful Soup and understand the challenges of parsing varied HTML structures across different recipe sites.