Interactive LEO constellation simulation for assessing a 1 GW space-based data center — Built for MAE 426

• Overview: A browser-based orbital mechanics simulator for designing and analyzing low Earth orbit satellite constellations. Explores the architecture behind a 1 GW solar-powered space data center by letting users tune constellation parameters and see real-time coverage, gap, and power metrics.
• Technologies: Three.js r128, Vanilla HTML/CSS/JS (no build step, no dependencies)
• Features: Live orbital simulation with adjustable planes, inclination, altitude, and phase offset; Starlink/OneWeb/Polar presets; coverage %, gap time, orbital period, and footprint area metrics computed client-side.

Autonomous track switching, obstacle detection, and train control — MAE 412 Capstone

• Overview: An embedded systems capstone simulating a segment of the Mount Rainier Scenic Railroad. The Vector Board Arduino autonomously manages track switches, detects fallen tree obstacles via ToF sensors, and forwards DCC speed/direction commands through a Python passthrough to a command station.
• Technologies: Arduino (C++), Python (serial bridge), VL53L0X ToF sensors (I2C), ACIA/SoftwareSerial, DCC, Fusion 360
• Features: Timed 3-pattern switch cycling triggered by train occupancy detection, dual-sensor obstacle detection with automatic track power cutoff, randomized servo-driven scenery animation, assembly firmware for signal sensor board.

Benchmarking HOG, Faster R-CNN, DETR, and SAM2 on the PnPLO Pedestrian Dataset — Princeton University

• Overview: A controlled architecture-level comparison of four pedestrian detection paradigms evaluated on the PnPLO (Person vs. Person-Like Objects) dataset, focusing on robustness to false positives from statues, mannequins, and other visually similar non-pedestrians.
• Technologies: Python, PyTorch, torchvision, OpenCV, NumPy, Matplotlib, Seaborn, Google Colab
• Features: Unified evaluation pipeline with precision, recall, F1, mAP, false positive breakdowns, PR curves, and confusion matrices across HOG+SVM, Faster R-CNN, DETR, and YOLO+SAM2.
• 📄 View Full Report

APPLE: Enabling Shap-E to run on Apple Silicon GPUs via Metal Performance Shaders (MPS) Acceleration

• Overview: Shap-E falls back to CPU on Apple M-series machines because certain indexing ops are not yet supported by PyTorch-MPS. This PR removes that blocker, giving native-GPU performance on macOS without sacrificing CUDA/CPU compatibility.
• Technologies: Python, PyTorch MPS
• Features: Significant performance increase while running Shap-E locally on Mac M-series, for example, on a Mac mini (M4 Pro), default image-to-3D generation time drops from 4 hours to just under 4 minutes when switching from CPU to GPU via MPS.
• 📄 View Pull Request

Tinder-style stock discovery powered by Claude AI — Built for the AI@Princeton x Trade[XYZ] Hackathon

• Overview: A mobile-first investing experience where users swipe right to invest and left to skip, with flippable cards showing live AI analysis, real-time news, and interactive charts.
• Technologies: React 18, Vite, Framer Motion, Express (Node.js), Anthropic SDK (Claude Haiku + Sonnet), Yahoo Finance, Finnhub API, YouTube Data API
• Features: Physics-based swipe gestures, Claude-generated stock hooks, AI portfolio analysis with Risk Radar, gamified streaks & badges, social friends feed & leaderboard, vertical Reels feed.
• 🎥 Watch Demo

An AI-powered compassionate medicine companion for elderly patients — Built for HackPrinceton F25

• Overview: NANI ("grandmother" in Hindi) is a full-stack IoT medication adherence system that uses IR beam sensors to passively detect when medication is taken and a voice-first AI assistant to answer health questions in multiple languages — keeping patients independent while keeping families informed.
• Technologies: Raspberry Pi (Python, FastAPI), Node.js (Express, OpenAI Whisper/GPT-4o-mini/TTS), Swift (iOS), Google Sheets (data logging)
• Features: Passive IR beam medication detection, bilingual voice assistant (English/Hindi), real-time Care Circle notifications for family and caregivers, cultural sensitivity, automatic adherence tracking.
• 🎥 Watch Demo

Using Frequentist and Bayesian Regression Models to Optimize Surprise Gift Strategies

• Overview: Developed a Bayesian Regression Discontinuity Design to evaluate and optimize the return on investment of surprise gift campaigns.
• Technologies: Python, NumPyro, Pandas, Jax, StatsModels, Matplotlib
• Features: Data Visualization & Actionable Insights.
• 📄 View Project README

Predicting Airbnb listing prices in New York City using various regression models.

• Overview: Developed models to predict Airbnb prices, with the Random Forest model showing the best performance.
• Technologies: Python, Jupyter Notebook, Scikit-learn, Pandas, Numpy, Seaborn
• Features: Data preprocessing, model training, evaluation metrics.
• 📄 View Project README

Modern and responsive website for Naacho Dance Company at Princeton.

• Overview: Features event details, member bios, and integrated Google Maps for location information.
• Technologies: HTML, CSS, Bootstrap
• Features: Responsive design, interactive elements, user-friendly interface.
• 📄 View Project README

Implementing CNNs and RNNs to analyze and predict trends in cognitive function.

• Overview: Proof of concept utilizing deep learning models for cognitive trend analysis.
• Technologies: Python, TensorFlow, Keras
• Features: Model implementation, data analysis, prediction visualization.
• 📄 View Project README

• LinkedIn 🔗
• Email 📧

✨ Thank you for visiting my profile! Let's build something amazing together. 🚀