This repository documents an experimental test of agentic AI workflows in academic research contexts. It serves as a proof-of-concept for how AI coding agents can assist in producing research artifacts.
The primary driver of this research is the human researcher. All AI-generated artifacts, documentation, code, and analysis contained in this repository are:
- Generated with AI assistance as part of a workflow experiment
- Subject to thorough human review and validation
- Intended to support, not replace, human research judgment
- To be rigorously verified before any formal academic use
This repository contains exploratory work and prototype artifacts. Any materials here that may eventually be used in academic publications will be:
- Extensively reviewed and rewritten by the human author
- Validated against source materials and ground truth
- Cited and attributed according to academic standards
- Submitted through proper peer review processes
webgsbench-app/- Benchmark viewer application prototypescripts/- Automation scripts for data processingtools/- Utility programs for scene conversion- Various documentation and planning files (experimental)
This project tested collaborative workflows between human researchers and AI coding agents (Claude, Kimi, etc.) for:
- Literature review organization
- Code prototyping and implementation
- Documentation drafting
- Experimental design planning
All AI-generated content requires human verification before use.
WebGSBench is a prototype benchmark framework for comparing 3D Gaussian Splatting formats and rendering approaches in web browsers. It includes:
- Multi-format scene support (.ply, .splat, .ksplat, .spz)
- Browser performance profiling
- Image quality metrics (PSNR, SSIM)
- Dual-viewer comparison interface
Status: Experimental / Proof-of-concept
Do not cite or reproduce materials from this repository without explicit permission from the author. This is an experimental workspace, not a finalized research publication.
Last updated: February 2026