Projects

Photo-first calorie logging with AI estimation and human confirmation.

Motivation: I built SnapFuel to reduce food logging friction while keeping the user in control. The core idea is: AI suggests, user confirms, dashboard reflects confirmed values.

Challenge: Model confidence needed transparent communication, not fake precision.

Research outreach copilot: scrape, summarize, draft, review, then send.

Motivation: LabReach started as a script to reduce repetitive outreach prep. I kept a human-review checkpoint as a hard requirement so automation helps without blindly sending.

Challenge: JS-heavy faculty pages broke simple scraping logic.

Dance practice tool with waveform selection and BPM analysis.

Motivation: Built from a dance practice pain point: finding the right section and tempo quickly. The app combines waveform interactions with server-side tempo analysis.

Challenge: Syncing waveform selection with backend analysis boundaries.

Bare-metal handheld Tetris on Arduino Nano + ST7735 display.

Motivation: I built this to practice embedded game loops and real hardware debugging. Everything runs on-device: rendering, input handling, collision checks, scoring, and sound cues.

Challenge: Intermittent display issues required solder-level debugging.

Computer vision dance coach with pose comparison feedback.

Motivation: This project explores whether movement feedback can feel useful instead of robotic. The pipeline compares user dance clips against references and surfaces actionable differences.

Challenge: Movement alignment is hard when tempo differs between performers.

Campus dining social app built through Penn SPARK.

Motivation: A real product-focused build with practical constraints, team collaboration, and user-oriented iteration. The goal was connecting students around dining in a way that felt lightweight and usable.

Challenge: Balancing feature ambition with limited project timeline.

This site, built as an evolving engineering journal.

Motivation: I treat this as a living system instead of a static portfolio. The project section itself now runs from reusable data so future additions are faster and more consistent.

Challenge: Keeping the site personal while scaling structure and consistency.

Remote vending controller that lets a user pick a bin in a Next.js dashboard and trigger a physical dispense through a Supabase command queue and Raspberry Pi worker.

Motivation: A cyber-physical vending pipeline instead of a dashboard-only app.; Frontend writes `device_commands` rows to Supabase with bin-specific actions (`dispense_bin_1..4`).; Raspberry Pi worker claims pending commands, executes servo movement script, then marks command status.; Servo control supports per-bin calibration; Bin 3 needed reduced duty cycle to prevent over-rotation.

Challenge: Reliable internet-triggered actuation with safe state transitions.