张皓霖 / Haolin (Harry) Zhang

科研工程师Research Engineer

西安交通大学 — 人工智能与机器人研究所Xi'an Jiaotong University — Institute of Artificial Intelligence and Robotics

参与文档编写、实验室共建、科研项目组织、工程开发与竞赛准备，指导低年级学生，并推动自动驾驶感知系统落地。

Contributed to documentation, lab co-construction, research project organization, engineering development, and competition preparation. Guided junior students and drove autonomous driving perception deployment.

技术合伙人 & 计算机视觉技术负责人Technical Partner & Computer Vision Technical Lead

杭州深想科技Hangzhou Shenxiang Technology (DeepThink)

作为公司核心合伙人，主要负责计算机视觉技术的研发管理，统筹生鲜识别、工业质检与后厨安全监控等核心项目的数据处理、算法设计和模型部署。带领团队与国隆电子合作研发支持增量学习的智能秤，形成国内及比利时量产规划；独立研发结合红外分割与半监督生成对抗学习的笔帽质检方法，在得力文具生产线实测达千分之三误报率。后厨系统上线百度 AI 市场，并应用于杭州高速服务站。

As a core partner, led computer vision R&D management, overseeing data preparation, algorithm design, and model deployment for produce recognition, industrial inspection, and kitchen safety. Led a team with Guolong Electronics on smart scales supporting incremental learning, with production plans for China and Belgium. Independently developed infrared segmentation and semi-supervised generative adversarial methods for pen-cap inspection, achieving a false-positive rate of 0.3% in tests on a Deli stationery production line. The kitchen system was listed on Baidu AI Market and deployed at Hangzhou highway service stations.

感知算法工程师Perception Algorithm Engineer

浙江大华技术 — 浙江华锐捷技术有限公司（原零跑汽车 ADAS 部门）Zhejiang Dahua Technology — Huaruijie (formerly Leapmotor ADAS Division)

开发多任务视觉训练框架并对接集群平台，预研结合变分自编码器与自监督优化的细粒度车型分类。参与 T03/C11 NOA 交通信号、交通标志与地面标识识别，模型迭代两版后上车部署，投入应用并持续 OTA 升级；融合视频检测跟踪、伪标注和自动评测，约半年完成公司首版视觉算法自动化评测与数据标注系统，自动化标注/评测准确率＞95%。

Developed a multi-task vision training framework integrated with the cluster platform, and explored fine-grained vehicle classification using variational autoencoders and self-supervised optimization. Contributed to T03/C11 NOA traffic-light, traffic-sign, and road-marking recognition, deployed on vehicles after two iterations and put into use with ongoing OTA updates. Built the company’s first automated evaluation and annotation system in about six months by combining video detection and tracking, pseudo-labels, and automated evaluation, achieving automated annotation/evaluation accuracy above 95%.