I build biological models you can explore and research tools you can reuse. My work connects computational biology, neural simulation, and AI-assisted research, with an emphasis on inspectable assumptions and reproducible workflows.
Two interactive habitats, from a crawling larva to a flying adult. Open a demo, follow the animal, and edit its world.
A detailed GLB fruit fly with ground exploration, autonomous 3D flight, an editable habitat, and a third-person camera. Three.js · MaleCNS model · Lightweight flight |
A deformable larva connecting neural activity, segmental movement, and an editable world, with virtual neuron-silencing experiments. Larval connectome · MuJoCo · Three.js |
Browser demos use reduced controllers for instant access. Full neural models and MuJoCo physics run locally; each repository documents its data sources and modeling limits.
Keep one ChatGPT conversation in the loop while Codex checks proposals, applies files, and verifies work in the repository. JavaScript · Agent Skill · Local MCP |
Turn papers, slides, and research materials into editable scientific posters matched to the venue, with rendered visual checks. Codex · Claude Code · OpenClaw · PPTX |
More research workflows: independent model review · advisor & lab due diligence · Challenge Class posters.
| Project | Question / approach |
|---|---|
| Bridge UMI Error Correction | Bayesian, region-aware, indel-tolerant correction of molecular identifiers. |
| Protein Language Models for DNA Binding | Protein embeddings, classical ML, CNNs, and LoRA for DNA-binding prediction. |
| Kuramoto Brain Network Modeling | Connectome synchrony, network hubs, perturbations, and disease-network simulations. |
| Biology Model Evaluation | Evidence-based evaluation of biological models through case studies. |
Working toolkit — Python / NumPy / SciPy · JavaScript / Three.js · MuJoCo · C++ · Git / GitHub · agent workflows.
Current interests — connectome-informed behavior, embodied biological models, protein representations, and research tools that make their evidence easy to inspect.
Data products, coursework & earlier experiments
- LinkVerse — creator discovery through data collection, potential scoring, and multimodal matching.
- Glimmer Scout — creator-potential and resonance analysis with GBDT and multimodal models.
- Bioinformatics · Peking · 2026 Spring
- Programming Practice 2026
- PKU iGEM Exercises
Open a project's issue tracker to discuss its methods or contribute. Browse all repositories →