Stanford EE + CS student | prev. iCloud Privacy Platform @ Apple
Currently: in Madrid, interested in mechanistic interpretability, training for Boston Marathon qualifiers
A COOL Project: My Custom COOL Compiler ->
Stanford / Bay Area · SF Marathon ->
Apple (Software Engineering Intern, 2026): Built out a token refund mechanism for the iCloud Privacy Platform.
Aida (SWE Intern, 2025): Built Python/SQL/GCP backend services for CRM automation and production workflows.
Stanford AIMI (Software Intern, 2022-2023): Trained a ResNet-50 chest X-ray device detection pipeline to 97% accuracy.
Melcher Lab, Stanford (Software Engineer, 2020-2025): Co-developed a Jetson + iOS + GCP liver analysis system with 100x lower latency.
CS143: Compilers (building a compiler from scratch!)
CS111: Operating Systems
CS231n: Deep Learning for Computer Vision
CS229: Machine Learning
CS234: Reinforcement Learning
EE108: Digital System Design
Papers, clinical computer vision, and course research projects.
5 items
Compared RLHF, PPO, DPO, and likelihood-regularized DPOP across robotic manipulation benchmarks, finding that standard DPO can destabilize competent continuous-control policies while RLHF and DPOP are more robust.
Preference optimization, RLHF, DPO, DPOP, LIBERO, Meta-World, RoboSuite, RoboCasaEdge-deployed liver biopsy analysis with tiled U-Net segmentation and on-device steatosis scoring.
Python, U-Net, TensorFlow, OpenCV, Jetson NanoTrained device detection and localization models for radiology workflows using chest X-ray data.
PyTorch, Multi-label classifier, YOLOv5, Clinical CVBenchmarked centralized training against independent PPO for coordinated autonomous vehicle platoons.
Stanford CS229, Multi-agent RL, autonomous vehicles, Flow, SUMOBuilt a shape-safe Rust deep learning stack with compile-time tensor guarantees and Apple Silicon acceleration.
Stanford CS231n, Rust, Apple Silicon, const generics, MetalApps, robotics, hardware, and course builds beyond the featured work.
11 projects
An app for city recommendations
Swift, Firebase, iOS developmentRecreating Anthropic’s dictionary learning approach to extracting interpretable features from a language model.
Sparse Autoencoders, Dictionary Learning, TransformersBuilt a custom COOL compiler for Stanford CS143, covering lexical analysis, parsing, semantic analysis, code generation, and backend optimizations.
C++, Compilers, MIPS, Stanford CS143, SystemsEE108 final project with chord synthesis, ADSR, echo, stereo panning, and VGA note display.
Verilog, FPGA, Audio DSP, Vivado, ROM sequencing, VGAConverted static intake into a conversational widget with Discord handoff and AI follow-up.
Vanilla JS widget, Discord.js, Mistral API, WebhooksLarge-format lighting and projection structure with mapped geometry and LED routing.
Blender mapping, Teensy firmware, LED simulation, ProjectionUsed iPhone LiDAR and 3D reconstruction to generate custom-fit insole geometry from scans.
Swift, ARKit LiDAR, Open3D, Trimesh, OpenCVMapped hand gestures to Nintendo Switch controls with low-latency tracking on Jetson.
YOLOv8, MediaPipe, OpenCV ORB, Jetson Orin, nxbt HIDHigh-altitude payload for telemetry, sensor logging, and cutdown control during balloon flight.
Dual Arduino, Sensor telemetry, Cutdown control, Python analysisVision-based pose estimation and heading correction for closed-loop autonomous navigation.
OpenCV, Blob tracking, Pose correction, Closed-loop controlPrototype for automatic solar-to-battery switching and stable dual-source power delivery.
Dual-source power switching, Motor transients, Capacitor tuning