仰政兴

Zhengxing Yang

邮箱：e@yzx.ai Email：e@yzx.ai

双专业：机器人工程（全美本科专业排名前五）、计算机科学Double Major: Robotics Engineering (Top 5 in U.S. undergraduate programs), Computer Science
专利：2项中国专利（CN102880944A、CN108770170A）+ 1项美国临时专利（节能减排装置）Patents: 2 Chinese patents (CN102880944A, CN108770170A) + 1 U.S. provisional patent (Energy saving & emission reduction device)
状态：休学创业，因疫情及公司经营未返校Status: On leave from university for entrepreneurship; did not return due to COVID-19 and company operations

训练并离线部署行业垂类LLM，满足私有化部署与合规要求Trained and deployed industry-specific LLM offline to meet private deployment and compliance requirements
设计开发多Agent联动框架，实现任务分解、规划与执行流程自动化Designed and developed multi-Agent collaboration framework for automated task decomposition, planning and execution
构建Agent友好图数据库，实现多实体双向关系、实体消歧与合并算法Built Agent-friendly graph database with bidirectional entity relationships, entity disambiguation and merging algorithms
实现实体关系 + Embedding + BM25 多参数混合优化的深度搜索能力Implemented deep search capability with multi-parameter hybrid optimization using entity relationships + Embedding + BM25

离线部署多个大语言模型与多模态Embedding模型并开发相关应用Offline deployed multiple LLMs and multimodal Embedding models and developed related applications
根据应用场景微调LLM，精选数据集完成训练（CPT + SFT）Fine-tuned LLMs based on application scenarios using carefully selected datasets (CPT + SFT)
部署向量数据库完成多模态数据分类及抽取Deployed vector database for multimodal data classification and extraction
实现多Agent联动的任务分割、计划与执行流程Implemented task splitting, planning and execution with multi-Agent collaboration

作为项目发起人，与美国新能源汽车企业中国分公司达成合作As project initiator, partnered with a U.S. new energy vehicle company's China branch
主导车联网边缘计算终端及设备主控项目全流程Led the full project lifecycle of IoT edge computing terminals and device main control systems
带领团队完成：商务接洽、需求分析、研发计划、工业设计、PCB设计、嵌入式开发、云端架构、CI/CD、DevOps、微服务中台Led team through: business negotiation, requirements analysis, R&D planning, industrial design, PCB design, embedded development, cloud architecture, CI/CD, DevOps, microservices platform
助力客户完成与中国某应急安全研究机构、中国某清洁能源研究机构的项目落地Helped client complete project implementation with Chinese emergency safety research institution and clean energy research institution

为40万平米、10万人次展会（每年3场，共约25万人次）提供观众实时画像与兴趣分析Provided real-time visitor profiling and interest analysis for 400,000 sqm, 100,000-person exhibitions (3 events/year, ~250,000 visitors total)
通过小程序推送个性化观展路线，IoT屏幕实时投放定制广告Pushed personalized exhibition routes via mini-program; delivered targeted ads on IoT screens in real-time
推动展后交易量提升7.85%，成为展会数字化行业标杆Increased post-exhibition transaction volume by 7.85%, becoming a benchmark in exhibition digitalization

整合无人机、地面站与自动移动平台Integrated drones, ground stations and autonomous mobile platforms
应用SLAM、点云3D地图、三维导航、传感器融合、视觉定位、深度学习、行为识别、人脸识别等技术Applied SLAM, point cloud 3D mapping, 3D navigation, sensor fusion, visual positioning, deep learning, behavior recognition and face recognition technologies
实现园区日常无人机与无人车自动巡逻及异常预警Achieved automated drone and UGV patrol with anomaly detection and early warning

为大型跨国药企提供网安法及个人信息出境风险控制咨询Provided cybersecurity law and personal information cross-border transfer risk control consulting for a major multinational pharmaceutical company
与行业伙伴开发咨询事务所自动化处理系统，应用于其他四大咨询公司药品流向审计项目，显著降低人力成本Co-developed automated processing system for consulting firm, applied to drug flow audits for other Big Four firms, significantly reducing labor costs

与WPI教授团队研究加密货币交易方式及套利算法Researched cryptocurrency trading methods and arbitrage algorithms with WPI professor's team
开发自动化监控系统，实时追踪各国数字货币价差Developed automated monitoring system to track cryptocurrency price differences across countries in real-time
构建多货币风险对冲及交易系统，实现稳定被动收入Built multi-currency risk hedging and trading system, generating stable passive income

大模型：LLM: LLM部署、微调、多Agent系统、向量数据库LLM Deployment, Fine-tuning, Multi-Agent Systems, Vector Databases
嵌入式：Embedded: 工业设计、PCB设计、嵌入式程序开发Industrial Design, PCB Design, Embedded Development
云端：Cloud: 架构设计、CI/CD、DevOps、微服务中台Architecture Design, CI/CD, DevOps, Microservices
其他：Other: SLAM、点云处理、深度学习、行为/人脸识别SLAM, Point Cloud Processing, Deep Learning, Behavior/Face Recognition