Mingjian Li | Portfolio

Crafting digital experiences with code and creativity.
Specializing in 3D Reconstruction, Machine Learning, and Full-Stack Development.

Explore highlights

Highlights

3D reconstruction

DanceNet3D

A multi-view dance dataset with 3D Gaussian Splatting reconstructions and AR/VR playback.

CVPR 2026 3DMV Workshop · Best Paper Award

View research

Robot learning & 3D geometry

World Action Models

Integrating scene geometry into robotic action prediction through parameter-efficient adapters.

FastWAM · VGGT-Ω · LIBERO

View research

Medical imaging

MRI Reconstruction

Reconstructing high-quality MR images from undersampled data with deep learning.

FastMRI · Variational networks · Diffusion models

View research

Beyond research

Broadway

A seat in the theatre: favorite shows, collected Playbills, and memories from Broadway.

All-time favorite · Hadestown

Explore Broadway

李明键 / Mingjian Li

A little about me.

Hi! I am Mingjian Li (李明键).

I am an MSE student in Computer and Information Science at the University of Pennsylvania. I graduated from NYU Tandon with a B.S. in Computer Science in May 2026. I am passionate about and have experience in 3D Reconstruction, MRI with Deep Learning, Full-stack Development, and Machine Learning. I have also developed a growing research interest in Robotics — from building robot arms to sim-to-real learning pipelines.

Before transferring to NYU, I studied Accounting at SUFE. Outside of tech, I love cats, Broadway shows, hiking, photography, board games, and ping-pong!

LinkedIn ↗GitHub ↗Google Scholar ↗Email ↗

Education

Aug 2026 – Present

University of Pennsylvania

MSE in Computer and Information Science

Graduate School of Engineering and Applied Science

2024 - 2026

New York University

B.S. in Computer Science; Minor in Mathematics

GPA: 3.972/4.0 · Dean's List

NextGenPhD Scholars Program
$5,000 research grant for 3D reconstruction and medical imaging (Fall 2025)

Undergraduate Summer Research Program (UGSRP)
$5,000 grant for research on 3D Gaussian Splatting (Summer 2025)

Coursework & grades

Machine Learning: 4.0
Machine Learning Systems: 3.7
Software Engineering: 4.0
AR/VR: 4.0
Image and Video Processing: 4.0
Vertically Integrated Projects: P
Operating System: 4.0
Design and Analysis of Algorithms: 4.0
Object Oriented Programming: 4.0
Data Structure: 4.0
Probability and Statistics: 4.0
Honors Numerical Analysis: 4.0
Computer Architecture: 4.0
Intro to Database: 4.0
Ordinary Differential Equation: 4.0
Discrete Math: 4.0
Calculus III: 4.0

2022 - 2023

Shanghai Univ. of Finance & Economics

B.S. in Accounting; Minor in Statistics

GPA: 3.71/4.0

Renmin Scholarship First Prize
Mathematical Contest in Modeling (2nd Prize)

Experience

Research, engineering, and everything in between.

Dec 2025 – Feb 2026
Chengdu, China

Zhiyuan Company

Robotics Software Engineer Intern

Built vehicle detection and high-precision localization module for automated parking system, fusing RGB camera data with Livox Mid-360 LiDAR point cloud to estimate 3D vehicle position and orientation.
Developed modular ROS-based software with YOLO real-time detection, achieving ±5cm localization accuracy in dynamic parking environments.

ROS · LiDAR · YOLO · C++ · Python · SLAM · Point Cloud

Dec 2024 – May 2025
Pittsburgh

SeeM(useums)

Software Engineer Intern

Frontend: React, Next.js, Redux. Built authentication, project management.
Backend: Python Flask, Docker, AWS. Scalable deployment.
3D Pipeline: Implemented 3D Gaussian splatting & SuGaR with CUDA.

React · Next.js · Python · AWS · 3DGS · CUDA

Sep 2023 – Jan 2024
Remote

WorldQuant Brain

Quant Research Consultant

Built market-neutral strategy based on cross-field data and sentiment.
Refined strategy to improve Sharpe ratio, turnover, and fitness.

Quantitative Research · Backtesting · Pandas · Financial Modeling

Jan 2023 – Feb 2023
Shanghai, China

KPMG

Audit Intern

Conducted risk assessment and substantive procedures.
Verified integrity of assets, payroll, and expense accounts.

Audit · Risk Analysis

Research

September 2026 – Present

Integrating World Action Models with 3D Geometry

Advisors: Prof. Lingjie Liu and Chen Wang (PhD student)

3D feature fusion. Researching geometry-conditioned models of the current scene for robotic manipulation policies, built on FastWAM and VGGT-Ω. Integrating point clouds or scene features extracted from dual-view RGB images into action prediction, with an end-to-end workflow for feature caching, adapter training, and closed-loop simulation.
Parameter-efficient training. Implemented point-cloud and scene-feature adapters that map geometry into 32 conditioning tokens. With the geometry encoder and policy backbone frozen, trained adapters with 2.38 million and 4.30 million parameters using flow matching loss, and completed a comparative study of the two geometric representations.
Ablation evaluation. Completed 350 closed-loop evaluations across seven configurations and 10 LIBERO tasks, comparing real, shuffled, and zeroed geometry inputs, as well as an adapter-disabled baseline. Used paired action losses and stage-by-stage timing to evaluate the contribution of geometry and its online computational overhead.

Publications

"DanceNet3D: A 3D Dance Dataset with Multi-View Videos and 3DGS Reconstructions."

Shihang Wei*, Mingjian Li*, Ran Gong, Yueyu Hu, Yao Wang.

CVPR 2026 3DMV Workshop · Best Paper Award
* Equal contribution

Project Page ↗Google Scholar ↗

NYU Video Lab: 3D Dancing Human Reconstruction Dataset

Computer Vision · 3DGS · PyTorch · Camera Calibration · Unity AR/VR · Unreal Engine

A benchmark dataset with multiview, multiframe human dancer sequences, including building a full pipeline for raw data processing, 3D Gaussian construction with PSNR ≈ 34.

Unity player integrated with AR/VR capabilities for real-time 3D Gaussian Splatting visualization, enabling interactive playback on mobile devices with seamless cross-platform deployment.

Unreal Engine plugin with native VR support for immersive 3D reconstruction visualization, featuring high-performance rendering and interactive VR display capabilities for enhanced spatial understanding.

View DanceNet3D Dataset ↗

3D Reconstruction Examples

AttitudePromenade

BartSimpson

BiancaGolden Chimee

Chacha

HouseFootworkAdvanced

RobertRubama RussiaCostume

Scroll to explore all six sequences. Use the video controls to pause.

Unity AR/VR Display

A Unity plugin for streaming and playing back animated 3D Gaussian Splatting sequences in real time, with AR and VR support.

Real-time sequence playback with transport controls
Play, Pause, Loop modes with adjustable speed (0.1x - 5x)
Batch conversion of PLY files to optimized assets
VR/AR support for Quest 3, Vive, Varjo and more
Interactive timeline and frame-by-frame scrubbing

View GitHub Repository ↗

Unreal Engine Display

A production-ready Unreal Engine plugin for streaming and rendering animated 3D Gaussian Splatting sequences with native VR integration.

Memory-efficient streaming with bounded VRAM (60MB for unlimited lengths)
Built-in playback UI with keyboard controls (P, comma, period)
Full transport controls: Play, Pause, Stop, frame scrubbing
Multi-sequence playlist with runtime switching
Native VR support with immersive display capabilities

Deep Learning Based Accelerated MR Image Reconstruction

Deep Learning · MRI · PyTorch · CUDA · FastMRI · Diffusion Models

GitHub Repository ↗Research Report ↗Presentation ↗

Research on deep learning–based magnetic resonance (MR) image reconstruction, aiming to accelerate scan times by reconstructing high-quality images from undersampled k-space data.

Comprehensive evaluation study of supervised, self-supervised, and zero-shot learning approaches using CNN, UNet, and diffusion-based models (Variational Networks, Score-based Diffusion Models) on the FastMRI dataset.

Technical Achievements & Results

VarNet Implementation: Achieved PSNR of 34.62 ± 3.78 dB with 4× acceleration and 0.24s inference time per slice
Score-Based Diffusion: Implemented SENSE parallel imaging achieving 12.5× speedup (from 30,000 to 2,000 evaluations)
Multi-GPU HPC: Refined PyTorch/CUDA implementations with distributed training achieving 4× speedup in performance
Systematic Evaluation: Tested 7 checkpoints to identify optimal model (epoch 50) with best generalization performance

Image Reconstruction Pipeline

Undersampled MRI — **Undersampled**Low-quality input

Deep Learning ModelUNet / Diffusion

Recovered MRI — **Recovered**High-quality output

Projects & experiments

From low-level kernels to full-stack systems.

Robotics · Spring 2026

SO-ARM101 Robot Learning Pipeline

3D printed and assembled an SO-ARM101 robotic arm, calibrated Feetech servo motors, and configured leader-follower teleoperation. Building a robot learning pipeline with teleoperation data collection for policy training and sim-to-real experimentation in Isaac Lab.

Robotics · Isaac Lab · Python · Teleoperation · Sim-to-Real · Servo Control

ML Systems · Fall 2025

ML Systems Implementation

Implemented GPU tensor operators (CUDA/C++) and MLP with auto-differentiation via pybind11. Achieved 97% MNIST accuracy with optimized kernels on NVIDIA T4 GPUs.

CUDA · C++ · Python · PyTorch · pybind11

↗

Full-Stack · Spring 2025

SkyRoute: Airline Management System

Cloud-based airline platform with flight scheduling, booking, and admin control. Normalized MySQL (3NF) on AWS RDS with Next.js RESTful APIs and CI/CD on Vercel.

Next.js · MySQL · AWS RDS · Vercel · REST API

↗

Full-Stack + AI · Oct 2024 - Present

Portfolio Management with AI

Portfolio management app with visualization dashboard for tracking holdings and market trends. Integrated LLMs and News API for sentiment analysis and tailored financial advice.

BERT · Transformer · OpenAI · React · MongoDB · Express

ML/AI · Oct 2024 - Present

Kaggle: Jane Street Forecasting

Real-time market data forecasting with data cleaning, PCA, clustering. Implemented SVM, CNN, RNN, LSTM, ARIMA with multiple sampling methods and regularizations.

Python · TensorFlow · Scikit-learn · Pandas · ARIMA

↗

Full-Stack + AI · June 2024

EcoPantry: AI Food Manager

Food management app with user authentication and AI chatbox for health analysis and recipe suggestions. Features dashboard with search and MongoDB storage for user data.

React · Express · MongoDB · Node.js · Azure API · Redux · Sass

↗

ML/Optimization · Sep 2023

Supply Chain Optimization

Math modeling project optimizing pricing and restocking strategies for supermarket vegetables. Processed 80,000+ data points using Random Forest, ARIMA, and Particle Swarm Optimization.

2nd Prize - China Mathematical Contest

Python · NumPy · Pandas · Random Forest · ARIMA · PSO

Away from the keyboard