Nathan's Notes

Ex-Stanford dropout working on creating the shared autonomy for remotely teleoperatable humanoid robots. Interested in implementing machine learning models from scratch, developing creative hackathon projects, and speeding up model training/inference with FSDP, quantization methods, and low-level architecture design. Check out my blog, Github, LinkedIn, or shoot me an email.

Work History

Nyro Robotics (Founder Fellow at South Park Commons)

Mar 2025 - Present

• Founded company creating systems for VR teleoperation of humanoid robots as 1 of 30 founders accepted to competitive South Park Commons Founder Fellowship (twice yearly cohort).
• Secured LOI with Booster (creator of world’s most robust robots) and raised $2M+ in preseed extension, receiving $400K @ 7% and $600K uncapped MFN.
• Managed interns, contractors, and technical hires to scale development operations.
• Trained RL policies using teacher-student distillation for humanoid self-stabilization and lower body adaptability to upper body dynamic teleoperation, even while carrying heavy weights.
• Led deal-making and developed NREs with VR camera and broadcast companies for non-line of sight 4K stereo video transmission and VR camera depth stitching.
• Built VR headset control systems using Vuer with pedal control, enabling users to remotely walk robots while seeing through the robot’s eyes.
• Collaborated with companies developing tactile gloves and ICs for force detection to provide haptic feedback for enhanced teleoperability.

Stanford Artificial Intelligence Laboratory (SAIL)

Sep 2024 - Mar 2025

• Developing dynamics-guided diffusion policies for manipulation of deformable materials by leveraging a transformer as a dynamics model for latent autoencoder representations of camera point cloud.
• Generating data using a custom simulator built with Taichi for accurate and efficient physics-based simulations.
• Optimizing test-time FID results by applying gradients to guide diffusion policies, inspired by classifier guidance in diffusion models for image and video generation.

K-Scale Labs

May 2024 – Oct 2024

• Developed quantized OpenVLA for the open-source robot “Stompy” using bitsandbytes, focusing on CUDA integration at company with $2M+ in humanoid robot preorders.
• Led PPO and physics simulator development with MJX and JAX, exporting Flax models with Tensorflow.js for inference.
• Integrated quantization methods into PyTorch by coding at the Python/C++ interface with Pybind.

Stanford University School of Medicine

Sep 2023 - Dec 2024

• Developed deep learning models with Pytorch and Weights & Biases to identify Right Ventricular insertion points for AHA Segmentation model on cDTI MRI images.
• Created an annotation system using AWS S3 for efficient data management.
• 4th in lab’s team for MICCAI 2024 MRI Reconstruction Challenge. Utilized variable splitting with CG method for data consistency, adaptive algorithm unrolling according to sample rate, and prompt blocks according to sample distribution to reconstruct undersampled MRI images.

Jane Street

Mar 2024

• Achieved 1st place in an intensive quantitative trading program by developing high-frequency arbitrage trading bots.
• Conducted manual arbitrage strategies with ETFs to identify real-time opportunities and maximize returns.
• Built a deep understanding of arbitrage mechanics and market dynamics in a live-trading environment.

University of Delaware

Feb 2021 - Dec 2022

• Led an independent investigation on phosphorene’s anisotropic properties, publishing results on ChemRxiv and presenting at conferences including but not limited to the American Physics Society and Society of Engineering Science.
• Conducted quantum simulations (LAMMPS, SIESTA) and analyzed DFT data, modeling stress-strain responses and identifying novel anisotropic material applications.
• Published conference proceedings, demonstrating applications in flexible electronics, catalysis, and fracture-resistant materials.

The Summer Science Program

Jun 2022 - Jul 2022

• Designed procedures and characterized UmCdc14 phosphatase protein using wet-lab methods and MOE drug-discovery software.
• Conducted homolog modeling, ligand docking, BLAST sequencing, and other computational techniques for enzyme inhibitor design.
• Published data into Purdue’s research repository after writing a comprehensive report.

Education

Stanford University, School of Engineering

Sep. 2023 – May 2026

• Relevant Coursework: Computer Organization & Systems, Polya Problem Solving Seminar, Modern Mathematics: Discrete Methods, Introduction to Probability Theory, Applied Matrix Theory, Machine Learning, Design and Analysis of Algorithms, Deep Learning for Computer Vision, Natural Language Processing with Deep Learning
• Clubs: Stanford Daily, BASES, Undergraduate Research Program, Student Robotics Club
• Research Conferences: American Physics Society March Meeting, MIT Undergraduate Technology Research Conference, Society of Engineering Science Annual Technical Meeting, Sigma Xi Student Research Conference

Skills

• Languages: Python, Typescript, C/C++, Java, JavaScript, Swift
• Frameworks: React, React Native, Django, NodeJS, Pytorch, TensorFlow
• Technologies: Firebase/Firestore, GCP, AWS, Docker, Git, Redis, Apache Kafka, Electron, XCode, Android Studio, CUDA
• Awards: Google Company Challenge Winner @ LAHacks, Telemedicine & Health Winner @ HopHacks 2024, 3x AIME Qualifier, 2x USNCO Finalist (National Top 50), USACO Gold (International Top 2000), 2x USABO Semifinalist (National Top 10%), Coca-Cola Scholarship Finalist