We are an AI solution provider focusing on the sales of conversational AI products and solutions for daily life and healthcare related application scenarios in China. Our business centers on empowering customers across various industries with user-centric AI solutions that improve operational efficiency, enhance decision-making and deliver better outcomes. These solutions are built on our proprietary central technology platform, UniBrain, which serves as the foundation of a wide range of applications, from intelligent customer service and healthcare diagnostics to multilingual conversational assistants. Our technology leverages continuous human interaction to refine its capabilities. In 2023, we introduced UniGPT, a 60 billion-parameter large language model with generative multilingual and multimodal capabilities, which enables human-machine interaction through dynamic reasoning and cross-modal data processing. Supported by our Atlas AI infrastructure, we ensure efficient model training and deployment, driving iteration and industry-specific adaptation.
We have offered products and solutions for a broad range of application scenarios in AI in Daily Life and AI in healthcare. China’s AI solution market is highly fragmented. According to Frost & Sullivan, we remained the fourth largest AI solution provider by revenue with a market share of 0.6% in China in each year of the Track Record Period. While the market size of AI solution in China increased by 33.1% from RMB135.5 billion in 2023 to RMB180.4 billion in 2024, our revenue increased by 29.1% from RMB727.3 million in 2023 to RMB939.0 million in 2024. In 2024, we ranked third by revenue in AI Solution in Daily Life and fourth in AI Solution in Healthcare in China. Extensive commercial application of these products and solutions has provided us with high-quality user feedback, which, in turn, has been prompting incessant iterations of UniGPT as the core of UniBrain.
AI Application Solutions
Sitting at the top of our technology stack, our AI application solutions layer is what users interact with. Based on UniBrain, we offer easy-to-deploy AI products and solutions with optimal performance that can be efficiently fine-tuned for a broad range of customers, helping them improve operational efficiency and productivity, as well as bring value to end users across industries. Specifically, the AI application solutions can be divided into Daily Life and Healthcare scenarios:
Daily Life: We provide diverse AI products and solutions that can be applied in application scenarios, including transportation, commercial space, hospitality and residential scenarios, among others, to improve the convenience and quality of people’s everyday life. Enterprises can provide immersive intelligent products and services empowered by our AI solutions to their customers, thereby improving their operational efficiency and service quality and reducing their management costs. For example, we developed the voice ticketing system applied in the auto ticketing machine of the Shenzhen Metro Line 20. Such system enables passengers to select their destinations through speech, shortening the average time spent on selecting a station on a ticketing machine from approximately 15 seconds in the traditional manual way to about 1.5 seconds. Such function enhances the efficiency of ticket purchasing and addresses the issue of congestion caused by queues at ticketing machines, thereby improving the overall passenger experience and optimizing the flow of traffic within the metro station. In 2022, 2023 and 2024, our Daily Life solutions empowered 299, 294 and 313 enterprise customers and 86, 112 and 121 system integrators/agents, which include a top three insurance group in China, Shenzhen Metro Line 20 and Xiamen Software Park. In addition, we provide AI large language model capabilities through MaaS to developers and enterprises on-demand. Key products include public cloud-based AI capability application programming interfaces (“APIs”) that enable different applications to communicate with each other, customized proprietary AI technology service platforms and AI model- embedded chips and Internet of things (IoT) hardware modules. In 2022, 2023 and 2024, we sold 12.8 million, 24.5 million and 36.0 million AI chips to developers and customers, respectively. We typically determine the final price of our offerings based on the total unit price of the products selected by the customer and the level of customization performed.
Healthcare: We offer AI-empowered healthcare solutions such as medical record voice entry, medical record quality control, single-disease quality control and medical insurance payment management. These AI solutions are typically delivered as customized AI-empowered business systems, which can regulate medical service processes and decisions, thus reducing medical errors, improving medical service quality and safeguarding the rights of the patients. For instance, in the context of medical record quality control, traditional information technology solutions, prior to the application of AI, could only perform basic quality control in terms of timeliness and completeness. However, with the application of natural language understanding and clinical knowledge mapping, it can now conduct quality control in terms of standardization and rationality, greatly enhancing the scope and depth of quality control and reaching a degree of completion close to that of manual expert quality control. Specifically, natural language understanding technology is used to identify entities and extract relationships from medical record texts. Through understanding these documents, it can model and recreate the changes in a hospitalized patient’s condition and the process of diagnosis and treatment. Then, the knowledge stored in clinical knowledge graph is applied for reasoning to determine whether the diagnosis and treatment actions reflected in the medical record documents are reasonable, thereby conducting in-depth quality control.
These AI solutions can also help medical staff increase efficiency. For example, our medical record voice entry system accelerates medical record entries by four to six times, compared to typing input; our medical record quality control system reduces the time required for quality review by 80%, compared to manual review, and thus is capable of a full inspection of all the medical records, compared to a sampling inspection of medical records if reviewed manually. They also help prevent unreasonable medical expenses, supporting medical reform and improving the utilization and efficiency of medical insurances. See “— Our Offerings — Healthcare.” In 2024, we provided AI solutions to 166 customers in Healthcare. We typically determine the final price based on the functionalities of the solutions provided and the level of customization performed.
Pricing
We have developed different pricing policies for our products and solutions. Specifically:
‧Daily Life: We generally provide customers with a menu-style quotation of our Daily Life solutions and products. The quotation of each solution typically depends on, among others, cost of sales, supply and demand, the marketing positioning, and the price of competitive solutions and products. We combine the selected solutions and products and make appropriate customized modification for our customers.
-Daily Life solutions, AI Chips and AI Modules: We typically determine the final price based on the total unit price of the products selected by the customer and the level of customization performed.
-AI Capability APIs: We charge subscription fees for such services on a pay-per-use basis.
During the Track Record Period, such subscription fees were insignificant to our total revenue.
‧Healthcare: Our Healthcare solutions are typically delivered as customized one-stop AI-empowered business systems, where a selection of our AI soft-and hardware are integrated according to specific customer demands. We also provide standardized efficiency tools, such as voice interaction solutions and service solutions, based on conversational AI capabilities. We generally determine the price of our Healthcare solutions considering various factors, including market positioning, market conditions, policy requirements, type of customers, pricing of competitors, functionalities of the solutions and level of customization.
Atlas AI Infrastructure
Our Atlas AI Infrastructure consists of the hardware and software resources which enable computing, storage, connectivity, scheduling and management. It supports the development, optimization and operation of our central technology platform, UniBrain, as well as UniBrain’s core algorithm model, UniGPT. We strategically started building the Atlas AI infrastructure in 2016, which centers on our intelligent computing cluster, dynamically scheduling strong computing power for efficient machine learning tasks. Atlas’s intelligent computing cluster currently harbors a computing power of over 184 PFLOPS and a storage capacity of over 10PB, both of which can be expanded without interrupting training tasks. Our intelligent computing cluster can efficiently and dynamically dispatch thousands of GPUs for parallel computing and seamlessly perform dynamic scale-up to address shifting business demand. It also optimizes the utilization of resources such as storage, bandwidth and computing power for large-scale machine learning tasks, addressing their significant demands for parallel computing. The strong computing power, efficient and dynamic dispatch and scalability of the intelligent computing cluster form our core competitive strengths in advancing AI solutions.
UniBrain
Featuring high universality, adaptablity and efficiency, UniBrain is the central technology platform which can be delivered through chips, edge-side SDKs and public and private cloud deployment. This central technology platform powers our AI application solutions, and is where our developers build, test and manage the AI application solutions. UniBrain comprises a core algorithm model and a wide array of AI components. The core algorithm model had been UniCore since 2019 until it was replaced by UniGPT in May 2023. UniCore was our first BERT-based large language model. Building on the evolution and advancement of the UniCore in past years, UniGPT features greater parameter and data scale as well as generative capabilities, and, pre-trained with massive volume of text and code, it has over 60 billion parameters and possesses strong general abilities.
Core Technologies and Applications
Large Language Model (LLM)
Our large language model UniGPT is the core of UniBrain, which determines the cognitive and business processing capabilities of the system and is the key to our MaaS-based business model. Interaction systems using traditional technologies often struggle to understand natural language expressions of users. For a well-calibrated system, even minor changes in user expression or the need to use contextual relations could easily lead to system failures. Large language models have remarkable advantages in understanding and generating natural languages. They enable a human machine interaction system to more easily understand varied user expressions, either in voice or text, and produce more human-like responses, offering accurate, easy-to-use and efficient user experience.
Industry-scale Knowledge Graphs
Industry knowledge graph is a core component for cognitive intelligence, which, when combined with UniGPT, can effectively improve the accuracy and reliability of industry knowledge and solve industry problems. Common knowledge about people’s lives, various rules and systems of architectural spaces and even different departments of science knowledge can all be structuralized and graphed. When performing human- machine interaction tasks, AI can refer to the content in the knowledge graph, allowing the system to provide reasonable, effective and specific responses. For example, location information of subway stations can be pre- stored in the knowledge graph. When a user inquires the location of the restroom, the system, after understanding the inquiry, can retrieve the information from the knowledge graph and then use natural language generation to produce text information and deliver such information to the user via voice response. Similarly, through the machine learning process performed on a great scale of medical knowledge, AI can develop systematic understanding of medicine and provide process management and decision-making support.
Multimodal Perception and Generation
Our multimodal perception and generation cover the chain of communication that starts from perception of external signals and ends with production of signals from response, facilitating the AI-enabled communication methods supported by our conversational AI core technology. This technology can be exemplified by the ticketing machines along the Shenzhen Metro Line 20. The machine uses computer vision to perceive whether there is a user in front. If a user is detected, the dialogue function is activated; if not, the machine remains silent.
Once activated, the machine uses a microphone array to capture the user’s voice, analyze the content of the speech, understand the user’s intent and provide response or perform the corresponding operation. This is a typical process of multimodal perception. During the dialogue with the user, the machine can perform certain operations, display results or text on the screen or provide hints and feedback to the user via voice. Voice electronic medical record entry is another typical voice input scenario. The machine perceives the doctor’s voice through the microphone and extracts the effective information from the voice to form text. AI, based on its understanding of the text, then offers suggestions to optimize the content.
One-stop IoT Interconnection System
We have developed a one-stop IoT interconnection system where all the relevant spaces, devices, users and services are interconnected for specific application scenarios to redefine the organic interactions among people, events and objects. This system is primarily used in Daily Life scenarios. Consider smart home as an example: the endpoints of this system are various IoT devices, including execution devices such as lights, air conditioners and fans, and sensors such as temperature meters, action detectors and smoke detectors. The center of this system is the IoT system, and the endpoints and center are inter-connected via the Internet, among others. The system also comprises various terminals that interact with end users and various business systems that deal with operators, if any. For example, a user feels too cold and says to the interaction terminal via voice that the temperature needs to be raised. The system understands the user’s intention and adjusts the temperature. Meanwhile, the temperature sensors at various locations continuously transmit data to the IoT center via the Internet. The AI center, according to the knowledge graph, determines that even if the temperature is adjusted according to the user’s requirements, the temperature in some areas may still be abnormal. Therefore, it automatically files an anomaly alert commanding a specific maintenance personnel to check and resolve the issue. See “Business - UniBrain - UniGPT - AI Components” for a detailed explanation on our core technologies and applications.
Source: Unisound (09678) Prospectus (IPO Date : 2025/06/20) |