本人具备扎实的算法竞赛背景，长期的信息学竞赛训练培养了我严谨的算法功底与计算思维，也塑造了「先厘清问题本质、再着手实现」的工作习惯。这一基础使我在面对新技术与新领域时能够迅速把握核心，并在正确性、时间复杂度与边界条件等层面保持高度的敏感与审慎。

在软件工程方向，本人以 TypeScript 为主要开发语言，熟悉基于 Next.js（App Router）与 React 的服务端渲染与前端交互体系，并配合 Tailwind CSS 构建界面；数据层采用 Prisma 与 PostgreSQL，坚持 schema 优先与规范化迁移；客户端以 Flutter / Dart 覆盖 Web、移动端与桌面端多平台。此外，亦涉及 Node.js 服务端开发、WebSocket 实时通信、Docker 容器化部署、持续集成、对象存储以及支付与消息推送等第三方服务集成。本人重视工程结构的规范性与可维护性，倾向于通过清晰的分层设计与可由工具强制的边界约束，配合自动化测试与持续集成，保障复杂系统在长期演进中的稳定性与可回归性。

在智能体与机器人方向，本人以 Python 为主要语言，熟练运用 PyTorch 进行深度学习模型训练，依托 MuJoCo 与 MJX 构建物理仿真环境，并通过强化学习（以 PPO 为主）训练运动控制策略，同时具备 Warp 高性能仿真与多 GPU 分布式训练的相关经验。研究兴趣集中于全身运动控制、仿真到真实环境的迁移，以及小样本条件下稳健策略的训练，相关工作需在数据组织、训练策略与领域建模等多个层面协同推进。

上述两个方向在本人工作中相互贯通：一面是对工程纪律与质量的长期坚持，另一面是对复杂学习问题的建模与优化能力。无论构建生产级系统，抑或收敛一个不稳定的训练过程，本人均以可解释性、可测试性与可复现性作为统一的行为准则。

I bring a strong background in competitive programming, cultivated through years of algorithm training. This foundation has given me rigorous algorithmic reasoning and computational thinking, as well as a working habit of fully clarifying a problem before beginning to implement. It allows me to grasp the essentials of new technologies quickly and to remain highly attentive to correctness, time complexity, and boundary conditions.

In software engineering, I work primarily in TypeScript, with proficiency in server-side rendering and frontend development built on Next.js (App Router) and React, together with Tailwind CSS for interface styling. For the data layer I use Prisma with PostgreSQL, adhering to a schema-first approach and disciplined migration management. On the client side I cover web, mobile, and desktop platforms with Flutter / Dart. Beyond these, my experience includes Node.js backend development, WebSocket real-time communication, Docker-based deployment, continuous integration, object storage, and integration with third-party services such as payment and messaging. I place a strong emphasis on the structure and maintainability of software, favoring clear layering, tool-enforced boundaries, automated testing, and continuous integration to keep complex systems stable and regressible over time.

In the field of intelligent agents and robotics, I use Python as my primary language, with proficiency in deep learning with PyTorch, physics simulation with MuJoCo and MJX, and reinforcement learning (principally PPO) for training motion-control policies. I also have experience with high-performance simulation through Warp and multi-GPU distributed training. My interests center on whole-body motion control, simulation-to-real transfer, and training robust policies under limited data, work that requires coordinating data organization, training strategy, and domain modeling.

These two areas inform each other in my practice: one is a sustained commitment to engineering discipline and quality, the other the ability to model and optimize complex learning problems. Whether building a production-grade system or converging an unstable training process, I hold myself to a consistent standard of explainability, testability, and reproducibility.