Harris Song
Summary
B.S. Computer Science @ UCLA. Software Engineer at Shopify. Research in 3D Vision-Language Models and embedded ML systems.
Education
- B.S. Computer Science2027University of California, Los Angeles (UCLA)GPA: 3.82Courses: OS (Threading/Processes in C++), GCP/AWS, Databases, Distributed Systems, Gerrit/Critique, Monorepo, Kubernetes
Work Experience
- Software Engineer Intern2026-01 - PresentShopifyWorking on Search (Index Serving)
- Improving Shopify's Catalog Search (Elasticsearch-based) performance in C++
- Approx. Nearest Neighbor benchmark indexing, searching 3×10⁵ queries/s for large-scale (1B+ items) catalogs
- Optimizing up ten-fold (from 5×10³ to 3×10⁵ q/s), working w/ ML Pipelines, GCP, Terraform, Kubernetes for deployment
- Software Engineering Intern2024-09 - 2024-12LeidosWorked on Sea Archer autonomous vessel systems
- Lightweight implementation of Sea Hunter that has already circumnavigated the world
- Converted server-grade Kubernetes DevOps to lightweight Docker, RHEL/Rocky 8, enabling autonomous startup/stop
- Debugged Ansible Playbooks and System C++ Architecture, built on NVIDIA-like edge hardware
- Software Engineering Intern2023-06 - 2023-08Office of Naval Research (US Navy)IED detection and defusal software
- Developed detection and defusal software for improvised explosive devices autonomously found in the Iraq War
- Used OpenCV, Python for camera-based bomb detection, unit/integration tests with hardware-in-the-loop for validation
- Modernized fleet of Turtlebots to NVIDIA Edge-AI Jetson TX2, Astra Orbbec 3D to ZED RGB-D camera
- Member of Technical Staff2024-09 - PresentLA HacksSouthern California's Largest Hackathon
- Worked on lahacks.com, apply.lahacks.com, managing UX and statistics features
- Built backend features supporting 6k+ applications through PostgreSQL, JWT Auth (Supabase) and Next.js (Vercel)
- Improved submission workflow, 2x application count (from 3k to 6k) through UX, performance and reliability
- Software Engineering Researcher2025-01 - PresentStructures Computer Lab, UCLAHiddenFruits: Embedded/ML fruit detection system
- Multi-modal CV running C on NVIDIA Jetson Orin Nano
- Developed Linux kernel drivers and C firmware for FLIR Lepton thermal sensors
- Implemented SPI/I2C protocol, UART, and boot loader (extlinux.conf, DTB) on Ubuntu 20.04/Linux Tegra
- ML Researcher2025-06 - 2025-09GRASP Laboratory, University of Pennsylvania3D Vision-Language Transformer Models
- Researching 3D Vision-Language Transformer Models, developing intuition for the 3D World through co-training
- Running 2k parallel simulations on NVIDIA HPC w/ Python, CUDA, PyTorch, Multi-Modal (text, 3D Voxels) architecture
- State-of-the-Art model more generalizable than previous discrete architectures, accepted into NeurIPS 2025 Workshop
Skills
Technologies
- C++
- Python
- OpenCV
- PyTorch
- CUDA
- Elasticsearch
- GCP
- AWS
- Terraform
- Kubernetes
- Docker
- PostgreSQL
- Next.js
- Supabase
Embedded
- Linux kernel
- NVIDIA Jetson
- SPI
- I2C
- UART
- FLIR Lepton
- Jetson Orin Nano
Publications
- Avi: A 3D Vision-Language Action Model Architecture generating Action from Volumetric Inference2025NeurIPS 2025 - Embodied World Models for Decision MakingHarris Song, Long Le. Project: https://action-volume-inference.github.io/
- Reconstruction Using the Invisible: Intuition from NIR and Metadata for Enhanced 3D Gaussian Splatting2025AAAI 2025Gyusam Chang, Tuan-Anh Vu, Vivek Alumootil, Harris Song, Deanna Pham, Sangpil Kim, M. Khalid Jawed
- Demonstration of Wireless Synchronization Methods in Autonomously Controlled Fleet of Drones2024
Portfolio
- EcoLens - Wolfram Research Award Winner2023PortfolioTop 20/115 teams. Management software for sorting trash. Connected IoT to ML pipeline. Vision-based model on ~450 images for recyclables.
- FIRST Robotics - Team Captain and Mentor2019Portfolio1st in Nation (7000+ teams) twice in 2021, 2023. Built hardware-in-the-loop software for World Champs. Multi-camera fault-tolerant systems, TensorFlow ML 97% confidence vision detection.