MINDS

MINDS Champions

Meet the leaders building Meaningful, Intelligent, Novel, and Deployable Solutions to transform businesses and society. 

MINDS are pioneers of innovation technology transformations

These organizations are leading the way in applying frontier technologies to reshape business models, unlock new value, and deliver tangible impact across industries.

49

Selected Transformations

30+

Countries

16

Sectors

6

Impact Areas

Case Studies

Learn more about each project from the MINDS Champions.

 Cohort: 1 Locations: San Francisco & Palo Alto, United States Application Title: AI-enabled discovery framework for neurodegenerative therapeutics Team Members: Stanley Prusiner (executive sponsor), Jack Hidary (executive sponsor), Gregory Merz, Nadia Harhen, Andrea Bortolato, Ben Shields, Kelly RichdaleThe challenge Developing new drug therapies is slow, costly and uncertain. On average, bringing a candidate from early research to clinical use takes more than a decade, with most candidates failing along the way. These challenges are especially acute for neurodegenerative diseases such as Alzheimer’s and Parkinson’s, where disease mechanisms are not fully understood and animal models often fail to predict human outcomes. Traditional drug discovery relies heavily on experimental screening. Researchers test hundreds of thousands of molecules in the laboratory to identify a small number of promising leads. Even with modern automation, this process is resource-intensive and constrained by the availability of experimental data. When data is limited or noisy, AI models trained only on historical results struggle to generalise and rarely uncover novel candidates. As a result, exploration of chemical space remains narrow, and many potentially useful compounds are never examined. The solution Physics-guided AI for data-scarce discovery The UCSF Institute for Neurodegenerative Diseases partnered with SandboxAQ to develop a discovery framework that combines AI with physics-based simulation. Rather than relying only on existing experimental labels, the system uses quantum chemistry and molecular physics models to simulate how molecules interact with biological targets. These simulations generate high-quality signals that guide AI ranking models, effectively creating reliable training data where experimental results are scarce. This approach allows the AI to evaluate many more compounds while maintaining scientific grounding, even when real-world data is limited. How it works Simulating molecular interactions at scale The framework uses large quantitative models to estimate how strongly a molecule might bind to a target protein, a key factor in drug effectiveness. These simulations act as a filter, allowing the system to score and rank compounds before laboratory testing begins. By combining simulation outputs with AI-based ranking, the system can explore millions of compounds computationally. Researchers then select a much smaller set of high-quality candidates for physical experiments. In practice, this shifts much of the early discovery effort from wet-lab trial cycles to in-silico evaluation, reducing time and experimental burden. How it differs from conventional approaches Reducing reliance on brute-force screening Conventional discovery pipelines depend on repeated laboratory testing to compensate for limited predictive insight. AI models trained only on past experiments tend to reinforce existing patterns and struggle to identify novel chemistry. The UCSF–SandboxAQ approach integrates physical understanding directly into AI workflows. Simulations provide structured guidance, allowing the system to reason about molecular behaviour rather than extrapolate only from prior data. This enables broader exploration while preserving interpretability. Laboratory testing remains essential, but it is applied more selectively and later in the process. Impact 5.6 million compounds screened computationally within weeks 36× reduction in experimental effort compared with traditional workflows 30× higher hit rates relative to conventional screening approaches Why it matters Expanding the search for new therapies For diseases where biological insight is limited and failure rates are high, reducing uncertainty early in discovery is critical. The collaboration between UCSF and SandboxAQ shows how physics-guided AI can help researchers explore chemical space more efficiently while lowering experimental cost. By shifting early discovery toward simulation-informed evaluation, the framework supports faster progress toward viable therapeutic candidates without increasing risk or reliance on exhaustive laboratory screening. 
USA | Health, Chemicals
AI-enabled discovery framework for neurodegenerative therapeutics
SandboxAQ, University of California, San Francisco (UCSF)
Cohort: 1 Location: London, United Kingdom Application Title: AI-powered digital asset management platform for rail operations and maintenance Team Members: Uchikado San (executive sponsor), Koji San, Nadia Mazzino The challenge Rail networks operate under increasing pressure. Urban growth, climate commitments and rising expectations for reliable public transport are driving demand, while budgets remain constrained and infrastructure continues to age. Much of a rail operator’s cost base is tied to maintaining physical assets such as tracks, rolling stock and signalling systems, many of which are designed to operate for decades. Traditionally, maintenance relies on two main approaches. Reactive maintenance addresses faults after failures occur, often resulting in service delays and unplanned interventions. Planned maintenance follows fixed schedules based on time or mileage, regardless of the actual condition of assets. While predictable, this approach can lead to unnecessary replacements and inefficient use of resources. Some operators have introduced predictive tools using individual sensors, but these systems usually monitor components in isolation. Without a network-wide view, it remains difficult to prioritise interventions or understand how issues in one area affect overall performance. Overhead lines can span hundreds of kilometres in both populated and rural areas. Infrastructure managers must work hard to maintain this infrastructure which can be subject to weather damage, for example, which can result in potential faults. Inspections rely heavily on visual checks by drivers or maintenance teams, which are labour intensive.The solution A unified platform for condition-based rail maintenance Hitachi Rail developed an AI-powered digital asset management platform to support a shift from schedule-based maintenance to condition-based decision-making. The platform integrates data from trains, signalling equipment and track infrastructure into a shared operational view. By analysing this data continuously, the system helps operators and maintenance teams identify emerging issues earlier and plan maintenance based on actual asset condition rather than fixed intervals. The objective is not to replace existing maintenance practices overnight, but to support more targeted interventions that reduce disruption and lower long-term costs. How it works From sensor data to operational insight Sensors installed on trains and along the network collect data on asset behaviour during normal operations. This includes vibration, temperature and electrical signals, depending on the component being monitored. Edge computing processes time-sensitive data close to the source, enabling fast detection of anomalies, while cloud-based analytics support system-wide analysis and forecasting. A digital representation of the rail environment is built to simulate how assets behave over time. This allows operators to test scenarios, anticipate degradation and plan responses more effectively. For example, wheel bearing maintenance previously relied on precautionary replacement. Thousands of bearings were changed at fixed intervals to avoid failure. With condition monitoring in place, replacements are now based on measured wear, reducing waste and downtime. The same approach applies to overhead line equipment. Image recognition and sensor data help identify faults before they escalate, reducing the need for manual inspection in difficult or hazardous locations. How it differs from conventional approaches Moving beyond isolated monitoring Conventional predictive maintenance tools focus on individual components and often operate as standalone systems. This limits their usefulness when decisions must be made across a complex network. Hitachi Rail’s platform integrates asset data across trains and infrastructure, allowing maintenance decisions to be prioritised at the network level. By combining local processing with central analysis, the system supports both rapid response and longer-term planning. This reduces reliance on blanket replacement strategies and reactive repairs. Impact The adoption of HMAX has delivered measurable value, including up to 15% reduction in maintenance costs and 15% lower energy consumption.Why it matters Supporting reliable rail operations at scale As rail networks expand and infrastructure continues to age, maintenance strategies need to become more precise without increasing operational risk. Hitachi Rail’s approach shows how integrated asset data and AI-supported analysis can help operators focus effort where it is most needed. By shifting maintenance decisions closer to actual asset condition, the platform supports more reliable service delivery while making better use of existing infrastructure and maintenance resources. 
United Kingdom | Engineering & Construction
AI-powered digital asset management platform for rail operations and maintenance
Hitachi
Cohort: 2 Location: Riyadh, Saudi Arabia Application Title: AI thermal imaging analytics for diabetic foot risk assessment  Team MembersAmplifai Health: Meshari F. Alwashmi, Mustafa Alghali, Omar Saleh Ministry of Health of Saudi Arabia: Abdullah Mohammed Arabe, Abdullah Al Soheimi, Hani Mohammed Badahdah The challenge Diabetic foot complications often develop without clear symptoms and can progress rapidly. In many cases, ulcers and infections are detected only after tissue damage has already occurred, increasing the risk of amputation. Globally, diabetes-related amputations occur at an estimated rate of one every 20 seconds and account for a significant share of diabetes-related healthcare costs. Beyond financial impact, these complications can severely affect mobility, independence and quality of life. Traditional screening methods rely largely on visual inspection and specialist assessment. While effective in advanced cases, these approaches are limited in their ability to detect early signs of inflammation and are difficult to scale. In many health systems, including Saudi Arabia’s, the number of trained specialists cannot keep pace with the growing number of patients requiring regular screening. The solution Thermal imaging supported by AI Amplifai Health has developed an AI-powered thermal imaging platform designed to identify early signs of inflammation and tissue stress before visible symptoms appear. Thermal imaging measures variations in skin temperature, which can indicate abnormal blood flow or inflammation associated with developing complications. AmplifAI’s system uses this data to support early risk assessment rather than relying solely on visual examination. The platform applies a multi-stage AI pipeline. Custom computer vision models handle image preprocessing, signal enhancement and region-of-interest segmentation. These outputs are then interpreted by fine-tuned large language models trained on clinical literature, which generate standardized clinical risk scores to support decision-making. How it works From image capture to clinical risk assessment The system is trained on more than 5,000 thermal images collected from hospitals, clinics, clinical trials and validated research datasets. Each image is linked to clinical data on ulcers, inflammation and related complications. Training data reflects variation in skin tones, environments and medical histories, supporting more consistent performance across patient populations. Screening results are delivered in under one minute, compared with approximately 15 minutes for traditional assessments. The platform can run on medical-grade devices and smartphones using edge AI, allowing operation without continuous internet connectivity. This enables deployment in community clinics and other settings with limited infrastructure. By identifying patients at elevated risk months before ulcers form, the system supports earlier intervention and more targeted use of specialist care. How it differs from conventional approaches Scaling screening beyond specialist capacity Conventional diabetic foot screening depends heavily on specialist availability and visual assessment. AmplifAI’s approach allows trained nurses to conduct AI-assisted screenings, with specialists focusing on patients identified as higher risk. This model helps address workforce constraints by increasing screening capacity without removing clinical oversight. The platform has been validated through clinical trials and approved for clinical use by regulatory authorities in Saudi Arabia, Bahrain and the United Arab Emirates. Impact Screening time reduced from 15 minutes to under one minute 90% reduction in clinician time per screening Up to 80% reduction in treatment costs for patients 12× increase in screening capacity More than 1,000 thermal scans per month across clinical sites Approximately 10% of screened patients identified as high risk, with follow-up confirmation in multiple cases  Why it matters Supporting earlier intervention in diabetic care The partnership between the Ministry of Health of Saudi Arabia and AmplifAI illustrates how AI-enabled screening can support earlier identification of diabetic foot complications in health systems facing specialist shortages. By expanding screening capacity and enabling timely intervention, the approach helps reduce avoidable complications while allowing clinical expertise to be directed where it is most needed. This supports more sustainable diabetic care as patient numbers continue to grow. 
Saudi Arabia | Health
AI thermal imaging analytics for diabetic foot risk assessment
AmplifAI Health, Ministry of Health of Saudi Arabia
Cohort: 2 Location: Rueil-Malmaison, France Application Title: On-device AI for efficient room control including HVAC energy optimization and automated comfort management Team Members: Philippe Rambach, Chief AI Officer; Benjamin Norman, Senior Principal, AI Product Manager; Scott Brown, AI Engineering Senior Manager; Mutaz Mujad, Product Manager (HVAC) Smart Room Control; Mark Van Den Broeke, Principal Project Design LeaderThe challenge Buildings account for nearly 38% of global energy consumption, with heating, ventilation and air conditioning (HVAC) systems representing the largest share. HVAC systems regulate temperature, humidity and air quality, making them essential for occupant comfort, but they are also among the most energy-intensive components of the built environment. As cities expand, weather patterns become more extreme and expectations for indoor comfort increase, demand for HVAC continues to rise. Facility managers must balance comfort with energy efficiency, often across diverse building types and usage patterns. Traditional control systems rely on static schedules, fixed setpoints and manual adjustments. These systems respond slowly to changing conditions such as occupancy, weather or internal heat gains, leading to unnecessary energy use. Cloud-based optimisation tools can improve performance, but they depend on reliable connectivity and centralised data processing. In many buildings, particularly those with privacy, security or operational constraints, cloud reliance limits feasibility. The solution Room-level optimisation using on-device AI Schneider Electric developed an AI-enabled room controller that performs HVAC optimisation directly on the device, without relying on cloud connectivity. The controller runs a Dynamic HVAC Optimisation algorithm entirely at the edge. It continuously measures local conditions, including temperature, humidity and occupancy, and predicts how each space responds to heating or cooling. Based on these predictions, the system automatically adjusts setpoints and start–stop times to maintain comfort while reducing energy use. Each room or zone is treated as an independent system. Rather than applying a single building-wide model, the controller learns how individual spaces behave over time and adapts control strategies accordingly. How it works Local learning without pre-training The system uses unsupervised learning to build thermal models directly from time-series sensor data and local weather inputs. Because learning occurs on-device, there is no need for pre-trained models, historical data uploads or continuous connectivity. The AI runs on low-power microcontrollers and updates models within milliseconds, allowing real-time response to changing conditions. All computation remains local to the controller, giving building operators full control over data and ensuring privacy. The solution is designed as a plug-in upgrade and can be deployed across a wide range of HVAC equipment, including fan coil units and rooftop systems, without requiring additional hardware or software layers. How it differs from conventional approaches Decentralising control to the room level Conventional building management systems optimise HVAC performance at a central level, often using averaged data and static rules. Schneider Electric’s approach decentralises optimisation, allowing each room to be controlled based on its actual behaviour. By embedding intelligence directly into room controllers, the system avoids delays associated with central processing and reduces dependence on network availability. This makes it suitable for retrofitting existing buildings and for sites where connectivity or data governance is a concern. Impact 5–15% reduction in energy consumption within the first two weeks of deployment Indoor comfort maintained in line with ASHRAE 55, a widely used global standard for thermal comfort Site-level emissions reductions observed in pilot buildings Thousands of devices deployed across four pilot sites, including Schneider Electric offices and early adopter clients Fully edge-based operation supports data privacy and simplifies integration with existing systems  Why it matters Improving efficiency where energy is actually used HVAC efficiency is often discussed at the building or grid level, but control decisions ultimately occur room by room. Schneider Electric’s approach shows how on-device AI can support more responsive HVAC control without adding complexity or external dependencies. By reducing energy use while maintaining comfort, room-level optimisation offers a practical pathway to improving building efficiency as energy demand and climate pressures continue to grow. 
France | Energy
On-device AI for efficient room control including HVAC energy optimization and automated comfort management
Schneider Electric
Cohort: 2 Locations: Iki City, Japan & Kanagawa, Japan Application Title: AI agent systems for hospital operations, bed management and patient flow optimization Team Members: Yuki Doi, Ito Momoka, Takamichi Mitsutake  The challenge Japan’s healthcare system faces structural pressure. A majority of hospitals operate at a financial deficit, while staff shortages continue to grow due to demographic change. These pressures are especially acute in regional and rural areas, where hospitals must maintain high occupancy to remain viable. Hospital management in Japan is also shaped by complex regulatory rules. Medical fee reimbursements depend on detailed documentation and strict compliance with government standards. Errors or omissions can lead to fee refunds, creating financial risk. At the same time, bed allocation requires constant adjustment to balance patient needs with operational constraints. Much of this work is handled manually, relying on experienced staff and disconnected systems. These conditions make it difficult to improve efficiency without increasing workload. The solution AI agents embedded into hospital management workflows Social Medical Corporation Genshukai partnered with Fujitsu to apply AI to hospital operations where complexity is highest. Rather than deploying isolated tools, Fujitsu introduced a set of AI agents designed to work within existing hospital processes. Each agent supports a specific function, such as checking compliance with medical fee rules, assisting with document preparation, or supporting bed allocation decisions. The agents operate together, covering the patient journey from pre-admission through discharge and coordination with referral facilities. Genshukai integrated this approach into its operating model, allowing staff to use AI support as part of daily work rather than as a separate system. How it works Supporting decisions that staff already make Medical billing rules in Japan are detailed and frequently updated. One of the AI agents continuously references official fee schedules and notifications when records are created. It highlights missing requirements and suggests corrective actions, helping reduce compliance-related errors. Another agent supports bed allocation. Using mathematical optimisation, it evaluates placement options based on capacity and patient needs. This assists staff in maintaining occupancy without manual recalculation. Additional agents support discharge planning and coordination with external facilities by displaying relevant conditions and availability. Each agent focuses on a defined task. An orchestration layer combines their outputs so staff receive clear guidance rather than fragmented recommendations. How it differs from conventional approaches Reducing reliance on manual coordination Traditional hospital management systems in Japan address billing, bed control and referrals separately. Staff must reconcile information across tools and apply complex rules from memory or paper guidance. The Fujitsu–Genshukai approach embeds rule interpretation and optimisation directly into workflows. AI agents support decisions as they are made, rather than reviewing outcomes after the fact. This reduces dependence on individual expertise and lowers the risk of error without removing human oversight. Because Japan’s healthcare system operates under a common national framework, the same agent-based approach can be applied across institutions without redesigning core systems. Impact Approximately 10% in annual revenue uplift from improved operations Monthly medical fee refunds reduced from $7,000 to zero through compliance support Around 400 staff hours per month freed from manual regulatory and bed-control tasks Active use by hundreds of nurses, administrators and care staff Why it matters Supporting sustainable hospital operations The collaboration between Fujitsu and Genshukai shows how AI agents can support hospital management by reducing administrative burden and improving consistency in complex processes. By embedding AI into everyday workflows, the approach helps hospitals operate more reliably under financial and staffing constraints. This becomes increasingly important as healthcare systems seek ways to sustain services without adding pressure to frontline staff. 
Japan | Health
AI agent systems for hospital operations, bed management and patient flow optimization
Fujitsu, Social Medical Corporation Genshukai
Cohort: 1 Location: Shanghai, China Application Title: AI-driven coordination platform for flexible manufacturing Team Members: Yuxiang Zhou (executive sponsor), Yongqiang Ren, Xiaokang Pei, Yihang Cai, Qiuru XiangThe challenge Manufacturing capacity is often poorly matched to demand. Large factories tend to focus on high-volume orders with narrow margins. Smaller factories frequently have idle capacity but lack access to customers or the ability to handle product design and onboarding requirements. At the same time, demand has shifted. Emerging brands and online creators increasingly want small production runs that can move quickly from concept to market. Traditional manufacturing systems are not designed for this. Product development relies on manual CAD work, experienced engineers and costly moulding processes. Manufacturing Execution Systems (MES) are expensive to deploy and slow to adapt, particularly for small and medium-sized factories that still operate with limited digitisation. As a result, production cycles stretch over many months, capacity remains underused and factories struggle to respond to fast-moving demand. The solution A coordination layer that links demand, design and production Black Lake Technologies developed an AI-driven platform that connects demand sources with distributed manufacturing capacity. The platform acts as a coordination layer rather than a replacement factory system. It links order intake, product design and production scheduling within a single workflow. Incoming orders are matched to suitable production lines based on capacity and constraints. This matching process, which previously required manual configuration, is completed in seconds. The platform supports both large and small factories. It enables smaller producers to access new demand while allowing larger manufacturers to handle short production runs without reconfiguring existing systems. How it works From order to production with minimal manual setup When a new order is received, the system maps product requirements to factory capabilities. It can process structured data as well as CAD files. Where designs do not yet exist, AI tools generate manufacturable 3D models from text prompts, applying constraints that avoid features that are difficult to produce. Production tasks are coordinated through a lightweight, smartphone-based interface rather than traditional MES software. Role-specific AI agents guide scheduling and workflow execution, reducing reliance on manual planning. This allows factories to adjust quickly as orders change. The system is deployed across a hybrid architecture that supports rapid onboarding and local execution, making it suitable for factories with limited IT resources. How it differs from conventional approaches Reducing dependence on rigid MES systems Traditional MES platforms are costly and difficult to customise. Changes often require months of configuration work, which limits responsiveness. Black Lake’s approach replaces fixed workflows with adaptable coordination. Instead of encoding every process upfront, the platform uses AI agents to handle design generation and production planning dynamically. Custom workflows can be deployed in days rather than months. This allows factories to handle smaller orders and faster cycles without investing in new infrastructure. Impact Utilisation increased from 65% to 83% Production cycles reduced from 6–12 months to 1–3 months Minimum order size reduced from 10,000 units to 100 units 18% improvement in capacity utilisation across deployed sites Platform supports 330,000 daily active users and processes millions of production-related requests each day Why it matters Making manufacturing capacity more responsive Manufacturing systems built for scale often struggle with flexibility. Black Lake Technologies shows how AI-driven coordination can help bridge this gap by linking demand, design and production in real time. By lowering the cost of onboarding and reducing cycle times, the platform helps factories use existing capacity more effectively while supporting new business models based on smaller production runs and faster delivery. 
People's Republic of China | Supply Chain & Transportation
AI-driven coordination platform for flexible manufacturing
Black Lake Technologies
Cohort: 3Members: Qichao Hu (Executive Sponsor), Kang Xu, Yumin Zhang, Dan Hannah, Adam Atanas, Bokyung (Olivia) Kim Problem: Developing better batteries for electric vehicles and energy storage remains slow and expensive because scientists must navigate vast chemical design spaces while relying on fragmented research data and time-intensive laboratory testing. Traditional approaches explore only a small fraction of possible materials, limiting progress in improving performance, safety, charging speed, and lifespan. Approach: The company has developed an AI-driven materials discovery platform that integrates scientific literature, molecular simulations, laboratory data, and autonomous experimentation into a unified system. The platform uses AI agents and chemistry-focused models to search through hundreds of millions of molecules, predict properties, recommend promising battery materials, and continuously improve through feedback from automated lab testing and human experts. Result: Reduced electrolyte discovery timelines from approximately two years to three months and cut physical experimentation by up to 70%, helping accelerate R&D efficiency, reduce costs, and speed up the development of safer energy storage systems. 
USA | Chemicals, Energy Technologies
AI-Driven Materials Intelligence Platform for the Energy Transition
SES AI Corp., Molecular Universe
Cohort: 3Members: Zhigang Zhou, Zhiqing Wang, Zuofei Shi, Ke Li, Jing Yu, Chuqiao NiuProblem: Air-conditioner manufacturers face growing demand for customized products and higher quality standards, but Process design, Process implementation and Process optimization still depend heavily on fragmented data and manual work. This slows product launches, limits automation and creates avoidable quality lostApproach: The transformation connects Process design, Process implementation and Process optimization in one AI-enabled workflow. Simulations reduce the need for physical trials, software code and process documents can be generated automatically, vision and noise models inspect products across the line, and AI agents analyze faults and feed production lessons back into design.Result: Deployed factories achieved: 37% faster product development, 49% higher production efficiency per person, full AI quality inspection coverage, 30.4% fewer first-pass yield losses, and 83% faster anomaly resolution.
People's Republic of China | Energy
AI-enabled Transformation of the End-to-End Air Conditioner Production Process
Hisense Group, Hisense-Hitachi
Cohort: 3Members: Tianlan Shao, Ao Fu, Zhuo Han, Shengyuan MaProblem: Modern manufacturing lines must handle dozens of product variants with tolerances as tight as a quarter of a millimeter, often in unsafe environments. Traditional industrial robots work only when every part is presented in exactly the same position. Each new variant demands custom fixtures and reprogramming, leaving these tasks to manual operation – a challenge that becomes harder as production scales.Approach: An intelligence platform, combining AI-driven 3D vision, world models, and standardised software, turning any existing robotic arm into an adaptive, intelligent agent. The same universal platform fits onto more than 1,000 different robot models, allowing retrofitting to existing lines without bespoke engineering for each installation. This delivers an “eyes, brain, and grippers” system that let robots understand and react to real-world complexity, while eliminating fatigue and reducing workplace risk.Result: On the partnered assembly line, single-shift capacity rose by 50%, and labor efficiency improved by 30%. Hazardous operations were fully automated, keeping people out of harm’s way, and running the line with the flexibility to handle new variants without downtime or re-engineering. 
Germany, People's Republic of China | Engineering & Construction, Technology
AI Powered Intelligent Robot
Mech-Mind, KUKA
Cohort: 3Members: Linfeng Sun, Shuaizhou Wang, Jinyi Wang (Xuchang Pang Donglai), Richard Wong, Liang Tang, Feibai Zhu, Jingquan Yu, Shelley Hofer (Mettler Toledo) Problem: Supermarkets run up to three disconnected manual workflows within each station that do not talk to each other. In the back room, packaging machines require constant readjustment for every product, wasting film and increasing food-safety risks. At the fresh food counter, staff memorize thousands of products. At checkout, queues build up while cashiers look items up by hand. The result is higher costs, slower service, and low customer satisfaction.Approach: A cloud-edge AI system spans all three stations. AI-powered auto-tuning packaging machines, counter scales and checkouts each recognize products with 99% accuracy by combining camera images, weight readings and dimensional data, based on the algorithm purposely designed to resist contamination. A central layer monitors every device, and the cloud continuously retrains the shared model across all stores while filtering out faulty data from individual devices.Result: Checkout time per item dropped 60%, film waste dropped 20%, packaging pass rates rose 18 percentage points, and pre-pack machine downtime fell 30-60 minutes per store daily.
Switzerland, People's Republic of China | Food & Retail
Multimodal AI for Supermarket Operation Optimization
Mettler-Toledo (Changzhou) Measurement Technology Ltd, Xuchang Pang Donglai Trade Group Co. Ltd
Cohort: 3Members: (The Territorial Delegation of Educational Development, Vocational Training, University, Research and Innovation in Granada) María José Martín Gómez (Executive Sponsor) Rocío Linares Morera (The Mind, Brain and Behavior Research Center (CIMCYC) of the University of Granada (UGR)) Francisca Serrano (Executive Sponsor) Sara Mata Francisca Serrano (Executive Sponsor), Macarena de los Santos-Roig (ATENXIA) Manuel Hernandez (Executive Sponsor) Emilio JimenezProblem: Around 1 in 6 children experience attention and literacy-related difficulties, yet current intervention approaches for children and adolescents still rely heavily on standardized, clinician-led models and manual professional supervision.Approach: An AI-driven gamified platform that continuously adapts exercises and intervention pathways in real time, while keeping professionals in the loop.Result: Literacy-related cognitive outcomes improved by 25–35%, while the time required for monitoring and manual intervention adjustment decreased by up to 50% over a 16-week intervention period.
Spain | Education
AI-Driven Personalized Cognitive Intervention Platform
ATENXIA, CIMCYC, The Territorial Delegation of Educational Development, Vocational Training, University, Research and Innovation in Grana
Cohort: 3Members: Nikolaos Mitsakos, Patrick Bangert, Yuxing Ben, James Cassanelli, Ali Karimi, Vivianne Mahecha MartinezIssue: Before investing millions of dollars in drilling, companies like Occidental desire to better understand geologic reservoirs and where oil & gas are most likely to be found. Today, this process relies on geologists and engineers interpreting huge volumes of geological and seismic data, which takes time and can be difficult to scale. Small interpretation differences can impact drilling decisions, reserve estimates and capital allocation.Approach: The company developed a suite of AI tools that helps geologists and engineers create a shared first-pass interpretation of subsurface conditions. The system analyzes well logs, rock chemistry, sensor measurements, and decades of historical interpretations to identify matching rock layers, predict rock types before drilling, group areas with similar geological behavior, and estimate the presence of oil, gas and water underground. The tools are integrated directly into the software experts already use, allowing teams to review and refine the AI-generated results rather than replacing human decision-making.Result: Geological interpretation work that previously took months now takes days, with less drilling downtime, fewer drilling-tool changes, and reduced horizontal drilling time.
USA | Oil & Gas
AI Subsurface Characterization Investigation Intelligence suite (ASCII)
Occidental
Cohort: 3Members: Lu Bo, Ken Wong, Michael Yao, Simon Liu, Lin XinhuaProblem: As AI, enterprise and HPC workloads continue to grow rapidly, data centers are facing increasing pressure from rising energy consumption, constrained power availability and the need to scale computing capacity while reducing carbon emissions.Approach: Building an AI-driven orchestration system that continuously optimizes enterprise, AI and HPC workloads across cloud and data center infrastructure, transforming energy into a schedulable resource for next-generation computing.Powered by Lenovo xCloud AIOps, the solution introduces a unified full-stack decision engine spanning workload scheduling, compute resources, cloud infrastructure, cooling and power systems, enabling real-time, closed-loop optimization across the entire data center stack.  Result: Deployed and validated at the Shanghai Jiao Tong University data center, the solution achieved up to 11% reduction in energy consumption and up to 310 tonnes less CO₂ emissions annually, while enabling greater compute output and infrastructure efficiency under the same power budget.
People's Republic of China, USA | Education, Technology, IT and Communications (ICT)
Integrated AI Orchestration for Holistic Data Center Energy Efficiency
Lenovo, Shanghai Jiao Tong University
Cohort: 3Members: Xu Tao, Wang Qi, Li Jingxiang, Cai Dan, Yan Jiaxi, Wang YanProblem: In solar cell manufacturing, millions of ultra-thin silicon wafers continuously move through highly automated production lines operating 24/7. While wafer-level traceability systems can record more than 1,800 process parameters for each individual cell, the data is typically distributed across multiple independent systems. When anomalies occur, engineers must manually correlate data across different platforms. Additionally, conventional traceability systems remain largely passive. As a result, corrective actions are delayed, yield performance declines, and manufacturing costs increase.Approach: Based on a wafer-level full traceability data platform and AI-driven algorithms, the company developed a dual-link closed-loop collaborative system framework, enabling precise root cause identification down to abnormal process steps on specific equipment and continuous monitoring of real-time process parameter fluctuations and abnormalities detection through dynamic predictive models.Result: Reduced root cause investigation from three days to three minutes, significantly improving problem response and corrective action efficiency, while increasing per-capita production capacity by 7%. The system also achieved a 0.5% improvement in overall solar cell yield and a 0.1% increase in cell conversion efficiency. 
People's Republic of China | Energy
AI-Driven Anomaly Traceability System for Solar Cell Manufacturing
Tongwei
Cohort: 3Members: Sheng Yang, Jia Tang，Tao Chen ，Quanqi Chen, Zhen Dai,Xueqing SongProblem: Critical power transmission routes can become overloaded when renewable generation, extreme weather, or grid topology changes shift electricity flows. Previously, operators relied heavily on manual checks, on-site experience, and repeated calculations within short decision windows, which made responses slower, more conservative, and harder to scale.Approach: The system continuously reads minute-by-minute grid data, forecasts, and network conditions. A reinforcement learning agent recommends safe adjustments to generators, storage assets, or controllable loads. Each recommendation is checked against operating constraints, tested through fast simulation, and reviewed by human operators before execution.Result: At the Guangxi control center, severe overloads were reduced by approximately 90%. Response speed improved by roughly threefold, with strategies generated in seconds, while also delivering cost savings.
People's Republic of China | Energy
Proactive Control AI Agent for New - Type Power System
China Southern Power Grid
Cohort: 3Members: (Valterra Platinum) Steven Bolleurs, Mark Freed, (Dunia Innovation) Ahmed Ismail, Rishi Kumar, Vineeth NairProblem: Industrial catalyst and materials R&D remains slow, fragmented, and heavily dependent on manual experimentation, with limited comparable evidence across materials, conditions, and performance metrics. This is especially critical in emission control and industrial systems, where material performance directly affects cost, uptime, and large-scale value creation, with development cycles and cost exposure reaching up to USD 300M.Approach: a closed-loop autonomous platform that combines autonomous robotics, AI experiment selection, simulations, and real-time learning to accelerate discovery, delivering experimental capacity in one day that is equivalent to the weekly output of 15 scientists. Result: Increased industrial experimentation throughput up to by up to 5,500%, approximately 56x, reduced R&D timelines from months to weeks, and lowered operating costs from up to USD 2,000 to less than USD 300 per experiment. 
Germany, United Kingdom, South Africa | Chemicals, Materials
Autonomous Catalysts Discovery and Acceleration Platform
Dunia Innovations, Valterra Platinum
Cohort: 1 Location: Beijing, China Application Title: AI forecasting models for energy trading and market optimization Team Members: Chi Wang, Jiabin Lu, Tom Lu, Ray Pang, Wenyuan BaoThe challenge Electricity markets are highly sensitive to weather. Wind speed, cloud cover and temperature directly affect how much renewable energy is produced and how demand fluctuates. For energy traders, even small forecast errors can translate into financial risk, especially in short-term electricity spot markets. Traditionally, weather forecasts used in energy trading rely on physics-based numerical models. These models simulate atmospheric behaviour using complex equations and large computing resources. While accurate, they are slow to run and difficult to update frequently. This limits how often forecasts can be refreshed and how finely they can be resolved at the local level. As renewable generation grows, markets become more volatile. Traders need forecasts that are not only accurate but also fast enough to support daily decision-making across large portfolios. Conventional models struggle to meet this requirement at scale. The solution A transformer-based weather forecasting system for energy markets Horizon Power partnered with TerraQuanta to deploy an AI-driven weather forecasting system designed specifically for electricity market use. The system replaces traditional physics-based forecasting with a transformer-based model trained on more than four decades of historical weather data. For non-specialist readers, transformer models are a class of AI systems that learn patterns from large datasets and can generate predictions quickly once trained. In this case, the model learns how atmospheric conditions evolve over time without having to simulate every physical interaction explicitly. The forecasting system integrates global meteorological records with real-time observations, allowing it to produce short- and medium-range forecasts suitable for trading decisions. How it works From weather prediction to market insight The model ingests satellite observations and terrain data to generate high-resolution forecasts of wind conditions and solar radiation. These forecasts are updated several times a day and can resolve local effects that are important for renewable generation. Because the model runs much faster than conventional numerical simulations, forecasts can be produced with significantly lower computational cost. This allows traders to explore scenarios and update strategies more frequently. Forecast outputs are then linked to electricity price models. By understanding how weather patterns influence supply and demand, the system supports more informed trading decisions across Horizon Power’s portfolio. How it differs from conventional approaches Speed and resolution over heavy simulation Conventional weather forecasting relies on repeated physical simulations that consume large amounts of computing time. As a result, forecasts are expensive to generate and difficult to scale across different regions or time horizons. The TerraQuanta–Horizon Power system relies on learned representations of atmospheric behaviour rather than continuous physical simulation. This reduces computational overhead and enables faster forecasting at finer spatial resolution. The result is a practical forecasting tool that aligns better with the pace of electricity markets. Impact Forecasting efficiency improved by a factor of 50,000 compared with traditional methods Deployed across a 2,000 GWh annual electricity trading portfolio Why it matters Supporting decision-making in weather-driven markets As electricity systems incorporate more renewable energy, weather becomes a central driver of market behaviour. The collaboration between Horizon Power and TerraQuanta shows how AI-based forecasting can help traders respond more quickly to changing conditions. By combining long-term meteorological data with fast inference, the system supports more timely decisions without increasing computational burden. This approach becomes increasingly relevant as energy markets continue to evolve. 
People's Republic of China | Energy
AI forecasting models for energy trading and market optimization
TerraQuanta, Shandong Hongzhi Shuke Electric Power Co., Ltd.
Cohort: 1 Location: Seoul, South Korea Application Title: Ultra-efficient on-device AI for autonomous robotics Team Members: Lokwon Kim executive sponsor), Dong Jin Hyun (executive sponsor), Tim Park, Moon Sub JinThe challenge Autonomous robots are increasingly used in indoor environments such as hospitals, offices and logistics facilities. These settings require robots to navigate safely around people, respond to dynamic obstacles and operate continuously on battery power. Early robotic systems relied on manual delivery or simple rule-based automation. These approaches reduced labour but struggled in complex environments where conditions change frequently. More advanced AI techniques improved perception and navigation, but they introduced a new constraint: computation. Running modern AI models typically requires graphics processing units (GPUs) or cloud connectivity, both of which consume significant energy and generate heat. For battery-powered robots, this creates a bottleneck. Power-hungry hardware shortens operating time, raises costs and limits deployment at scale. Reliance on cloud processing also introduces latency and connectivity risks, which are problematic for real-time decision-making in indoor spaces. The solution Custom AI hardware designed for edge robotics Hyundai partnered with DEEPX to develop an on-device AI solution that enables real-time inference without relying on cloud computing or energy-intensive GPUs. The solution centres on custom AI semiconductors designed specifically for inference tasks. These chips are optimised to run perception and navigation models efficiently within tight power and thermal constraints. Rather than adapting general-purpose hardware, the system is built to match the needs of mobile robotics from the outset. DEEPX provides the semiconductor architecture and optimisation software, while Hyundai integrates the technology into autonomous robotic platforms for real-world operation. How it works Running AI directly on the robot AI models used for tasks such as obstacle detection or path planning are optimised using proprietary software so they can run efficiently on the custom chips. This optimisation preserves accuracy while reducing computational load. A software development kit supports model conversion and deployment, allowing engineers to move from training to on-device execution without redesigning applications. Once deployed, inference runs entirely on the robot, removing dependence on external servers. Because computation happens locally, robots can respond immediately to their environment while maintaining low energy consumption. This makes continuous operation feasible in indoor settings where charging opportunities are limited. How it differs from conventional approaches Moving away from GPUs and cloud reliance Conventional robotic AI systems depend on GPUs or cloud processing to handle complex models. These approaches increase power draw and introduce delays, especially when connectivity is unreliable. The Hyundai–DEEPX solution replaces general-purpose hardware with purpose-built AI accelerators. By running models directly on the device, the system reduces energy use and avoids latency associated with remote processing. This makes advanced AI practical for small, mobile robots operating indoors. Impact 240% performance increase compared with GPU-based systems Energy consumption reduced to one-eighth of GPU requirements Up to 90% reduction in total cost of ownership for deployed robots Why it matters Enabling practical autonomy at the edge As robotics expands beyond controlled environments, energy efficiency and real-time response become critical. The collaboration between Hyundai and DEEPX shows how AI hardware designed for edge use can support autonomous systems without the overhead of cloud infrastructure or high-power components. By lowering energy demands while maintaining performance, the approach supports wider deployment of autonomous robots in settings where reliability and cost control matter most. 
Republic of Korea | Technology
Ultra-efficient on-device AI for autonomous robotics
Hyundai Motor, DEEPX
Cohort: 1 Location: Beijing, China Application Title: AI infrastructure across over 20 financial business domains Team Members: Yang Longru (executive sponsor), Jin Haimin, Luo Yi, Gao Hongsheng, Liu Chengyan, Bao Xinyue The challenge Banks operate under strict regulatory requirements while managing complex processes that span credit assessment, risk control, customer engagement and capital allocation. Although AI technologies have advanced quickly, their adoption in large financial institutions has remained limited. Traditionally, AI in banking has been applied in scenario-specific ways. Examples include customer service chatbots or isolated risk models. These tools can improve individual tasks, but they do not integrate across full business processes. Infrastructure limitations, data sensitivity and the need for auditability make it difficult to deploy large models at scale. As a result, many AI initiatives remain experimental and fail to translate into sustained operational impact. For a bank with thousands of branches and hundreds of thousands of employees, this fragmentation limits the ability to use AI consistently across the organisation.  The solution A financial AI platform built as shared infrastructure ICBC addressed these constraints by developing AI as a foundational capability rather than a collection of individual applications. At the centre of this approach is a large language model trained specifically on financial data and banking workflows. The model is designed to support reasoning across documents, transactions and regulatory rules, rather than generating generic responses. ICBC built a centralised AI platform around this model, allowing specialised AI agents to be deployed rapidly across business units. This infrastructure supports more than 20 financial domains, including credit operations, risk management and financial markets. Instead of building separate systems for each function, teams can adapt the shared platform to their needs while operating within a common governance framework.  How it works From model training to daily use The financial language model is trained using a combination of supervised learning, fine-tuning and reinforcement techniques, with data drawn from compliant public sources and ICBC’s internal records. Data processing includes automated validation and safety checks to meet regulatory standards. A central control layer manages deployment and monitoring. This allows new AI agents to be released to branches within minutes rather than months. Staff interact with the system through existing tools, using AI to assist with analysis, documentation and decision support. Human review remains part of the process in areas where judgment or accountability is required. The underlying infrastructure includes large-scale computing resources that support continuous training and updates without interrupting operations.  How it differs from conventional approaches Moving beyond isolated financial AI tools Conventional banking AI solutions focus on individual use cases and are often developed independently. This leads to duplicated effort and inconsistent standards across the organisation. ICBC’s approach provides a shared platform with domain-specific intelligence embedded at the core. Rather than adapting generic models to financial tasks, the system is designed around banking requirements from the outset. This enables broader adoption while maintaining consistency, security and auditability. The same framework can be extended across branches and teams without rebuilding systems each time.  Impact More than 400,000 users, with around 20,000 staff using the platform daily Approximately 50,000 person-years of workload handled through AI-assisted processes  Why it matters Making AI usable at financial scale For large financial institutions, the challenge is not whether AI works in principle, but whether it can be deployed safely and consistently across the organisation. ICBC’s platform shows how AI can be treated as shared infrastructure rather than a collection of experiments. By embedding domain-specific models into daily workflows and aligning them with regulatory requirements, the approach supports broader use of AI while preserving control and accountability. 
People's Republic of China | Financial services
AI infrastructure across over 20 financial business domains
Industrial and Commercial Bank of China (ICBC)
Cohort: 1 Location: Rueil-Malmaison, France Application Title: AI-enabled microgrid systems for energy optimization and carbon reduction Team Members: Philippe RAMBACH, Chief AI Officer; Aditya CHANGAVALLI, Product Owner; Navdeep AHUJA, Microgrid Offer Management Director; Vanya IGNATOVA, Senior Principal, AI Product Manager; Jacques KLUSKA, Principal, Responsible AI Specialist The challenge Buildings are a major source of energy use and carbon emissions. Heating, cooling, lighting and electrical equipment account for a significant share of global electricity demand, particularly in commercial and industrial facilities. Managing this energy efficiently has become more difficult as buildings incorporate rooftop solar, battery storage and electric vehicle charging. Most building energy systems still rely on fixed schedules and rule-based controls. These systems operate according to predefined settings that change infrequently. They do not respond well to real-time conditions such as weather shifts, fluctuating electricity prices or changes in occupancy. When tariffs rise or grid conditions become unstable, facility managers often have limited visibility and must intervene manually. As energy systems become more distributed, these static approaches lead to higher costs and underuse of on-site renewable resources. They also limit the ability of organisations to reduce emissions consistently across large property portfolios.The solution AI-driven control for on-site microgrids Schneider Electric developed an AI-enabled platform designed to manage energy across facilities that combine local generation and flexible consumption. For readers unfamiliar with the term, a microgrid is a local energy system that can operate while connected to the main grid or independently if needed. It may include solar panels, batteries and controllable loads such as heating or charging equipment. Schneider Electric’s platform connects a cloud-based optimisation engine with a local edge controller installed at each site. Together, they coordinate how energy is generated, stored and consumed throughout the day, with the goal of reducing cost and emissions without disrupting operations. How it works Predictive optimisation with local execution At the core of the system is a model predictive control (MPC) engine. MPC is a control method that uses forecasts to decide how a system should behave over a short future horizon. In this case, the engine considers site-specific data such as weather forecasts, expected demand and electricity prices. Every few minutes, the optimiser recalculates an operating plan. It decides when to draw power from the grid, when to use on-site generation and when to charge or discharge batteries. These decisions are executed locally by the edge controller, allowing fast response even if connectivity is limited. The AI models update continuously using incoming data. This allows the system to adapt to changing conditions without manual tuning or custom engineering for each site. How it differs from conventional approaches Moving beyond static energy rules Traditional building energy management systems apply fixed heuristics that are updated infrequently. They react to conditions after changes occur and cannot anticipate future constraints. Schneider Electric’s approach treats each microgrid as a dynamic system. By forecasting conditions and adjusting control actions in a closed loop, the platform responds more quickly to changes in weather or pricing. Because the intelligence layer is shared across sites, deployments do not require bespoke configuration, which supports scaling across diverse facilities. Impact 14% reduction in energy consumption across deployed sites 28% reduction in CO₂ emissions per site per year across 97 locations Approximately $430,000 in annual net savings at a pilot site Why it matters Making distributed energy systems easier to manage As organisations add renewable generation and flexible loads, energy management becomes more complex. Schneider Electric’s platform shows how AI-based control can help facilities use existing assets more effectively. By shifting from static rules to predictive optimisation, the system supports lower emissions and reduced costs while maintaining operational stability across different types of sites. 
France | Energy
AI-enabled microgrid systems for energy optimization and carbon reduction
Schneider Electric

MINDS Champions

Loading

MINDS

A World Economic Forum Initiative
© 2026 All rights reserved
Privacy Policy & Terms of Service