Projects & Research
Research Projects
Agentic3DGS: Adaptive Agents for Robust Single-Image 3DGS
Research Challenge: How can we achieve robust single-image 3D Gaussian Splatting when the input image quality is unexpected or imperfect?
- Agentic Pipeline: Developed a LangChain/Gemini system that dynamically selects restoration modules based on image quality analysis.
- Key Insight: Discovered that popular diffusion generators (Zero123++) produce geometric inconsistencies that break COLMAP; transitioning to Stable Virtual Camera resolved this.
- Personal Growth: Leveraged this project to ramp up on 3D Generative AI, mastering Gaussian Splatting and Diffusion models from scratch.
Technologies: Python, LangChain, Google Gemini, 3DGS, Diffusion Models
HN-SoGCLR: Learning Contrastive Learning through Experimentation
Research Challenge: How do implementation details—like hard negative mining and optimizer stability—actually impact the performance of CLIP-style contrastive learning models?
- Experimental Pipeline: Experimented with existing SoGCLR and iSoGCLR frameworks, systematically varying loss functions and optimizers to test training stability.
- Hard Negative Mining: Investigated methods to emphasize "hard negatives" (confusing samples), finding that targeted sampling strategies significantly influence retrieval metrics.
- Personal Growth: Gained deep empirical intuition for self-supervised learning, mastering experiment tracking with Weights & Biases and learning that rigorous evaluation often outweighs architectural complexity.
Technologies: Python, PyTorch, CLIP, Weights & Biases (W&B), Contrastive Learning
Interactive Sequin Board: Iterative Prototyping and Physical Experimentation
Research Challenge: How can we navigate real-world constraints—like gravity and mechanical tolerances—to design and fabricate a functional interactive sequin mechanism?
- Iterative Design: Built and tested multiple iterations to explore sequin layouts, pin structures, and attachment methods, prioritizing fabrication consistency over final optimization.
- Physical Experimentation: Conducted measurement experiments on geometry and spacing to improve motion stability, and implemented servo-based actuation for controlled flipping.
- Personal Growth: Mastered the transition from digital sketches to functional physical prototypes, developing deep skills in CAD, 3D printing, and design thinking.
Technologies: CAD, 3D Printing, Physical Prototyping, Servo Motors, Design Thinking
Work Experience
Research Assistant AI/XR - Spatial Interaction & Interaction Design Lab
May 2025 - Current - College Station, TX
Developed an intelligent VR anatomy navigation system to enhance medical training through multimodal interaction.
- System Development: Developed a VR anatomy navigation system in Unity for Meta Quest 3, implementing multimodal interaction (Voice + UI) for hands-free control.
- AI Integration: Integrated Azure Speech-to-Text and LLMs (Claude, Gemini) to interpret anatomy-specific commands, improving recognition accuracy for medical terminology.
- Research Pipeline: Built a semantic parsing pipeline to extract actions and targets, benchmarking sentence similarity models against LLM-based approaches.
- Analytics: Engineered a modular logging framework to capture end-to-end latency, enabling data-driven comparisons between AI and Human-assisted performance.
Technologies: Unity, C#, Meta Quest 3, Azure Speech, Hugging Face, LLMs
Related Blogs:
Technical Product Manager - Develop for Good (AllGrow)
Oct 2024 - Feb 2025 - Remote
Led the development of a React Native MVP to connect students and teachers, targeting 1M+ users.
- Leadership: Led a team of 8 developers to ship a React Native MVP, managing weekly sprints and roadmap integration.
- Product Strategy: Integrated stakeholder feedback into feature requirements and designed a scalable architecture to support high user growth.
Technologies: React Native, Agile/Scrum, Product Management
Medical Device Product Development Intern - Alcon
May 2024 - Aug 2024 - Lake Forest, CA
Streamlined medical device project management and enhanced ORA system interfaces.
- Full Stack Development: Built a web application using Blazor, C# and .NET to streamline project tracking for medical devices.
- System Enhancement: Enhanced ORA system software with intuitive WPF interfaces, directly improving ease of use and operational accuracy.
- Documentation: Applied expertise in ORA and AnalyzOR to improve technical documentation and application processes.
Technologies: Blazor, C#, .NET, WPF, SQL
Undergraduate Research Student - UrbanResilience.AI Lab
May 2023 - May 2024 - College Station, TX
Analyzed lifestyle recovery impacts after Hurricane flood damage.
- Data Analysis: Developed a Python program to analyze and quantify lifestyle recovery impacts following natural disasters.
- Visualization: Processed large-scale mobility data using Pandas and GeoPandas to visualize vulnerable regions and aid recovery planning.
Technologies: Python, Pandas, Numpy, Stats, GeoPandas
Software Engineer Intern - Uhnder Inc.
May 2022 - Aug 2022 - Austin, TX
Improved chip reliability analysis for ADAS Lidar systems.
- Reliability Engineering: Built a Python analysis tool for BLR data in ADAS Lidar, delivering six key features to track chip reliability.
- Optimization: Achieved 60% faster data analysis by modeling with linear regression (NumPy/Pandas), validated on 2+ years of historical data.
Technologies: Python, Pandas, NumPy, Linear Regression