Alexander (Alex) Phu

AskAAPM

First author · submitted to AAPM 2027 Spring Clinical Conference and JACMP

A retrieval-augmented system that answers questions about AAPM Task Group Reports and cites the source passages. Submitted to the AAPM 2027 Spring Clinical Conference and the Journal of Applied Clinical Medical Physics.

What I built

• A question-answering system for AAPM Task Group Reports.
• Every answer cites the source passages it was based on.

How it works

1. Task Group report PDFs are ingested
2. Reports are split into passages that can be searched
3. A question pulls back the most relevant passages
4. A short answer is written from those passages
5. The answer is shown with quoted evidence and citations

Tools: RAG, FastAPI, Streamlit, FastMCP

Status

• Submitted to the AAPM 2027 Spring Clinical Conference.
• Submitted to the Journal of Applied Clinical Medical Physics.

View GitHub · Open it on the 3D desk

3D-Printed Radiation Shield

Contributor · submitted to Brachytherapy

An application that turns treatment-planning CT data into a printable STL shield and an updated DICOM RT Structure Set for gynecologic brachytherapy. Manuscript submitted to Brachytherapy.

Tools: Python, DICOM, STL, 3D printing

VISTA-3D Segmentation Pipeline

Lab project · 2025

Python and FastAPI around NVIDIA VISTA-3D in Docker.

KinetIQ

COMPUTER VISION · SPORT · 2026–

An AI-assisted swing analyzer that tracks body and racket motion from video, then turns it into timing and consistency feedback for players and coaches. Submitted to the 2026 Congressional App Challenge.

What I built

• An upload screen for four activities: tennis forehand, tennis serve, golf swing, and golf putt.
• Pose tracking drawn over the player's own video, so they can see what the analysis saw.
• An efficiency score with a skill level, plus how many frames were analyzed.
• A joint-by-joint comparison (trunk lean, elbow, shoulder, knee) that puts the player's angle next to a pro reference.
• Recommended drills with sets and reps, with the top one marked for the player.

How it works

1. Upload a video and pick the activity
2. Track the body's joints frame by frame
3. Measure the key joint angles
4. Compare them with a pro reference
5. Show a score, joint feedback, and drills

Tools: Computer vision, Biomechanics, Data viz, MediaPipe, React, FastAPI

Status

• Submitted to the 2026 Congressional App Challenge.

View GitHub · Open it on the 3D desk

ChessAI Assistant

COMPUTER VISION · AR · 2025–26

Custom YOLO models and OpenCV read the board from Ray-Ban Meta camera feeds. Stockfish then suggests moves and explains tactics as the game unfolds.

What I built

• Custom YOLO models that find each piece and label it with its color, type, and square, such as bR @ a8 for the black rook on a8.
• Board reading with OpenCV that turns the detected pieces into a full position, written as a FEN string.
• Stockfish analysis that suggests the next move for the side to play and explains tactics as the game unfolds.
• Two ways in: a camera pointed at a physical board (including Ray-Ban Meta camera feeds), or Chess.com through screen capture.

How it works

1. Frames come in from a camera or the screen
2. YOLO finds and labels each piece
3. OpenCV maps pieces to squares
4. The position is written as FEN
5. Stockfish suggests the next move

Tools: YOLO, OpenCV, Python, Stockfish

View GitHub · Open it on the 3D desk

Vital Tracker + BreathWork

SwiftUI and HealthKit apps.

Devpost

RV Park Comparison

Search and compare parks by amenities and pricing.

Open