The Ministry of Industry and Information Technology has recently released its 2025 list of typical AI application cases, with Zhejiang Province securing 14 entries across six categories.

In the technology infrastructure category, Alibaba Damo Academy (Hangzhou) Technology Co., Ltd. was recognized for its RynnBot embodied intelligence development platform. The platform tackles the long technology chain and high R&D barriers prevalent in embodied intelligence development. By leveraging cloud-device synergy, it achieves "zero GPU deployment" on the device side, allowing even a few-hundred-dollar 3D-printed teaching robotic arm to run mainstream embodied models like VLA. This enables small teams and individual developers to enter full-chain embodied intelligence development at near-zero hardware cost. To foster the industry ecosystem, the academy launched the "RynnBot SparkEdu" training camp, collaborating with 18 universities domestically and internationally, accumulating 17,000 developers and incubating over 100 innovative works. Developer practice spans the entire process of remote data collection, model training, and real-machine deployment, continuously building talent and innovation capacity for the robotics industry.

In the industry empowerment category, Yidao New Energy Technology Co., Ltd. was selected for its AI-empowered photovoltaic cell module lifecycle project. This initiative addresses digital transformation and carbon neutrality goals in PV manufacturing, building a vertically integrated intelligent factory system around digital R&D, intelligent production, lean operations, and high-end services. It covers the full lifecycle of PV product design, manufacturing, operations, and maintenance. The project innovatively combines mechanism models with data-driven modeling, paired with 5G IoT millisecond-level data collection, edge real-time inference, and hybrid cloud elastic computing. It has developed 13 core AI models specific to PV, achieving defect recognition accuracy of 0.05mm and equipment failure prediction accuracy of 92%. The system also integrates nine core business systems like MES and ERP, establishing an industrial big data center and process knowledge graph to support intelligent optimization, visual inspection, predictive maintenance, smart scheduling, and business decision-making.

Huaxin (Jiaxing) Intelligent Equipment Co., Ltd. was recognized for its AI-enabled intelligent material transport for integrated circuits. The project redefines AMHS systems in semiconductor factories, breaking the limitations of traditional logistics that only passively execute transport tasks. By integrating production, logistics, and equipment data, AI injects intelligent decision-making into core AMHS operations, enabling dynamic path optimization and real-time congestion prediction. This improves OHT scheduling efficiency by over 20% and integrates deeply with CIM and MES systems, significantly reducing empty travel rates and wait times. AMHS is evolving from a factory logistics artery into an intelligent agent with autonomous perception and continuous learning capabilities, shifting from passive execution to proactive optimization, supporting wafer fabs in achieving further capacity growth under full production conditions.

Beilong Precision Technology Co., Ltd. was selected for its intelligent mold design and manufacturing solution for high-end smartphone precision components. In response to industry pain points such as experience-dependent mold correction, long process debugging cycles, and low precision inspection efficiency, Beilong collaborated with Zhejiang University on industry-academia-research efforts. They built a three-in-one system of intelligent design correction, dynamic process control, and precision intelligent inspection, leveraging AI and optimization algorithms to close the data loop between R&D, testing, and production. This enables adaptive cavity correction, real-time parameter compensation, and automated dimension inspection. The project targets high-precision micro-injection molding scenarios for high-end smartphone components. Research data shows mold correction cycles and process debugging time can be reduced by 30% to 40%, labor costs cut by approximately 25%, injection molding yield improved to over 98%, and equipment utilization increased by about 20%. The project has formed a series of independent intellectual property rights, breaking import dependence. Its modular AI system demonstrates strong cross-scenario adaptability, offering a replicable intelligent manufacturing path for high-end industries like precision optics.

Zhejiang Zheda Zhongkong Information Industry Co., Ltd. was recognized for its smart station application based on the "ChengGui YiZhi" large model at Jingfang Station. Traditional subway station operations rely heavily on manual control, with high labor costs, slow emergency response, high energy consumption, and lagging passenger services. Existing station control systems are mostly at the L2 semi-automation stage, requiring upgrades through large models and AI technologies. This project focuses on scenario-based smart station operations, developing the ChengGui YiZhi large model product system and a smart station control system. It builds an architecture of "one industry large model, two intelligent devices, multiple business agents," constructing an L3-level "AI + reduced manpower" smart station operation and control system. Successfully deployed at Jingfang Station on Hangzhou Metro Line 4, the project advances stations from traditional control to full-scenario autonomy. Notable results include 37.5% improvement in personnel efficiency, 50% reduction in operational workload, 42% energy savings with daily savings of 930 kWh, and passenger service and safety perception accuracy exceeding 98%.

Ningbo Preh Joyson Automotive Electronics Co., Ltd. was selected for its "large model + agent" driven automotive electronics testing system. Addressing common industry challenges such as low testing efficiency, difficult knowledge accumulation, and insufficient complex scenario coverage, Preh Joyson developed an innovative testing system combining large models and intelligent agents. The system establishes a trinity of knowledge-guided generation, model-driven simulation, and intelligent closed-loop diagnosis, advancing automotive electronics testing from experience-driven to knowledge-driven approaches. The system has achieved large-scale application in automotive electronics R&D testing and quality verification, serving over 20 internationally renowned automakers and more than 30 domestic new energy vehicle and industry chain leading companies. It has supported over 1,000 product models, project versions, and test configurations, with test task executions exceeding one million. For a single vehicle model ECU test, the cycle time has been reduced from approximately three months to 15 days, compressing the cycle by about 83% and improving test efficiency sixfold. Through knowledge consolidation, template reuse, and rule constraints, the system reduces dependence on the individual experience of senior test personnel and improves organizational efficiency in complex project delivery.

China Mobile (Hangzhou) Information Technology Co., Ltd. was recognized for its "AI + audio-visual" platform enabling full-chain audio-visual content operations. This project addresses the long production cycles, high labor costs, inconsistent content standards, and insufficient personalization in traditional content operations. The platform serves mobile HD, smart speakers, and other scenarios, leveraging a pool of tens of millions of media posters and 1080P frame extraction materials. It integrates multimodal large models, RAG knowledge bases, vector retrieval, intelligent recommendation, and audio-video detection enhancement capabilities, supporting material retrieval, AI creation, highlight mining, 4K/8K super-resolution, content enhancement, and quality inspection. The platform also pioneers 7.1.4-channel immersive audio output for set-top boxes. By aggregating various vertical AI capabilities, the platform achieves hour-level fine-tuning and millisecond-level inference, creating an orchestrated, highly automated, intelligently scheduled content generation and editing platform that drives full-chain intelligent upgrade of content operations.

In the product application category, Youdao (Hangzhou) Intelligent Technology Co., Ltd. was selected for its AI Answer Pen SpaceX. Targeting after-school tutoring, homework assistance, and self-directed learning scenarios across K12, the product relies on the self-developed "Zi Yue" education large model and the DeepSeek-R1 reasoning model as dual engines, combined with multimodal OCR recognition, on-device AI deployment, and TTS emotional voice interaction. It addresses pain points such as insufficient parental tutoring capability, students' lack of problem-solving logic, low recognition accuracy in traditional learning hardware, and limited interaction modes. With a focus on AI full-subject tutoring and intelligent learning loops, it provides solutions for nine subjects, whiteboard-style video explanations, step-by-step reasoning, interactive follow-up questioning, and intelligent mistake management. The product features a 4.4-inch low blue light eye-protection screen, an 8-megapixel AI camera, and a 3.6cm ultra-wide scan window, achieving 96% recognition accuracy for complex images and text. It supports both 4G and WiFi connectivity. Since launch, users average over 10 questions answered daily, and mistake organizing efficiency has improved from hours to minutes. The product has been piloted in multiple public schools with the first AI answer room established.

Zhejiang Shizai Intelligent Technology Co., Ltd. was recognized for its Shizai Agent general intelligent agent platform aimed at accelerating digital transformation for industrial enterprises. Addressing pain points such as complex legacy system integration, insufficient intelligent automation, low data utilization, and heavy reliance on manual decision-making, the platform uses domestically produced computing power and multimodal large model technology. It supports both SaaS and private deployment with security and control. Without requiring code or API modifications, it significantly shortens development cycles and reduces transformation costs, enabling multi-agent collaboration with high-speed response, high accuracy, and autonomous learning optimization. The platform seamlessly integrates with ERP and MES industrial systems, covering core business scenarios including production monitoring, finance automation, business data analysis, intelligent office workflows, and efficient warehousing management. It builds a data-driven intelligent decision system, breaking down technical barriers to digital transformation and enhancing overall operational efficiency and competitiveness.

Hangzhou Sanhai Electronic Technology Co., Ltd. was selected for its FaultsMind equipment fault diagnosis generative agent. The agent focuses on engineering pain points in high-end equipment intelligent maintenance, including heavy expert dependence, high customization requirements, long R&D cycles, difficult updates, and insufficient cross-scenario generalization. Tasks that traditionally require professional designers and continuous consumption of human and material resources, such as customized diagnostic algorithm development, regular diagnostic system maintenance, online diagnostic model updates, and configuration of diagnostic systems for new equipment, are fully replaced by FaultsMind. This shifts the diagnostic capability paradigm from "manual development and maintenance" to "autonomous generation by large models." FaultsMind possesses full-process capabilities from diagnostic requirement modeling to integrated verification and application deployment, with the ability to coexist and integrate with existing digital operations and maintenance systems. It is applicable to intelligent inspection, anomaly monitoring, defect identification and tracing, and fault diagnosis in high-end manufacturing, transportation, power energy, and aerospace sectors.

In the support and assurance category, Alipay (Hangzhou) Digital Services Technology Co., Ltd. was recognized for its NovaFlow intelligent agent benchmark evaluation and monitoring platform. Developed by Alipay, this application targets typical social scenarios such as transportation, government services, and employment. Built on three core capabilities, industry knowledge-driven hierarchical benchmarks, multimodal role-driven evaluation, and mixed-expert intelligent attribution, it elevates agent evaluation from simple question-and-answer correctness to user task completion effectiveness. The platform achieves minute-level online performance degradation detection and rapid intervention, compressing single-round offline evaluation cycles from about 14 days to three days. Currently, the platform covers evaluation needs across nine major transportation scenarios, over 16,000 government service items, and precise recommendations for more than 18 million jobs. It has empowered the deployment of over 80 government agents and more than 100 transportation agents, providing a replicable evaluation paradigm for agents moving from general Q&A to real-world task execution.

In the special topics category, SUPCON Technology Co., Ltd. was selected for its TPT time-series large model enabling autonomous operation in process industries. TPT is the first time-series large model for the process industry, serving as SUPCON's core industrial AI foundation. Built on high-quality industrial datasets, advanced algorithms, and superior computing power, it integrates simulation, control, optimization, prediction, and evaluation capabilities across the full lifecycle of factory design, operation, and maintenance. It significantly reduces the workload of production management and operations personnel while enhancing device operation safety, achieving 30% to 50% improvements in personnel efficiency. The model also improves production efficiency, reduces costs, stabilizes product quality, and maximizes overall device benefits, with expected efficiency gains of 1% to 3%. TPT has been validated in over 100 industrial cases, delivering significant energy-saving, carbon reduction, and economic benefits while ensuring production safety.

Hangzhou Anheng Information Technology Co., Ltd. was recognized for its Hengnao security intelligent agent empowering cybersecurity intelligence advancement. With the increasing frequency of AI-driven cyberattacks and traditional security solutions relying on signature databases and manual monitoring, pain points include alert redundancy, scarce expert resources, slow security implementation, and high operations costs. Anheng Information created the Hengnao security agent platform, based on a security-domain large model, offering two service models: mature agent sales and joint customer co-creation. Serving major event security teams, government and enterprise clients, and security service providers, the platform supports zero or low-code agent construction, unified scheduling of security tools, and consolidation of security operations experience. It has been deployed at the Asian Games, Asian Winter Games, financial institutions, and universities, serving over 500 clients with more than 150 commercial implementations. The platform automatically analyzes massive alerts, rapidly handles threats, consolidates security knowledge, shortens protection plan implementation cycles, reduces costs, and helps achieve zero-incident security goals with broad industry promotion prospects.

Wenzhou University Big Data and Information Technology Research Institute was selected for its "mode-data resonance" AI-empowered integrated platform for industry enterprises. The platform focuses on deep integration of model capabilities and industrial data, addressing complex needs in enterprise digital-intelligent transformation. It builds an empowerment system of "sensing, diagnosis, resource coordination, decision-making, and scenario reshaping." Leveraging self-developed intelligent applications like DeepServe agents, the platform transforms enterprise data, knowledge, and business rules into understandable, inferable, and executable intelligent capabilities. It enables AI to move from a single-point tool to deep integration throughout production and operations, continuously identifying problems, generating strategies, and driving optimization. This gradually forms a "smart brain" with enterprise cognition, decision-making, and evolution capabilities, advancing enterprise digital-intelligent transformation from passive response to proactive sensing, intelligent decision-making, and continuous optimization, addressing practical challenges of transformation.

These selected cases highlight Zhejiang's comprehensive strength across the AI industry chain, from foundational technologies to industry applications and security assurance, reflecting the region's role as a national leader in artificial intelligence innovation and application.