CS undergraduate at Dalian Ocean University (2023–2027).
Focusing on Computer Vision (monocular geometric pose estimation, crowd density regression) and AI Systems (LLM multi-agent simulation, hybrid RAG pipelines, and containerized sandbox runtimes).
- Personal Homepage: tianyasky.github.io
- GitHub: @TianyaSKY
- Contact: tianyaguest@qq.com
| Project | Domain | Architecture & Methodology |
|---|---|---|
| SKYCloud | AI Workspace & Agent Runtime | Integrates file storage, asynchronous RAG embedding pipeline, and isolated Docker OpenCode runtimes for supervised agent tool-use and code execution. |
| CrowedSigmod | Crowd Density Estimation | YOLO-PGMD: Introduces localized re-normalization for strictly mass-conserved Gaussian density labels ( |
| SKYCity | Multi-Agent Simulation | A 2D tile world built with PixiJS and FastAPI, featuring autonomous NPC planning, memory retrieval, and state-machine-driven environment interaction. |
| SKYOJ | Online Judge & Sandbox | Multi-mode code evaluation platform (ACM / OOP / Kaggle) with network-isolated Docker execution sandbox and asynchronous Celery worker queues. |
| RacingCar | Reinforcement Learning | Custom Gymnasium 3D racing environment with 16-dimensional observation, 9-ray obstacle sensing, and Stable-Baselines3 PPO parallel sampling. |
