Know You Like an Old Friend: Being Labs Develops Multimodal Models for Individual Life Forms
With the increasing popularity of wearable devices, smart cameras and various types of sensors, individual living beings are generating an unprecedented volume of data. However, most of these data only stay at the "recording" level, and it is difficult to convert them into continuous understanding of individual status and risk prediction. Being Labs is precisely targeting this pain point, trying to use the multimodal large model to enable machines to continuously understand the status changes of each individual living being just like a friend who has been together for a long time.
Being Labs is a multimodal model company oriented to individual living beings. Its core technical route is to integrate video, audio, sensor and long-term historical data, continuously track individual status, identify abnormal changes and predict subsequent risks. The company has long been targeting the human health, elderly care, life sciences and animal health markets, and serves equipment manufacturers and professional institutions with its model capabilities.
Being Labs plans to raise 10 million RMB in this round of financing to cover 12 months of operation, of which about 80% will be invested in technology R&D and 20% in commercial expansion. The first 6 months will focus on completing the prototype and early warning Demo, and the next 6 months will promote the first version of the model and paid pilots.
From "Group Statistics" to "Individual Status", Multimodal Models Cut into Life Health
Most of the current AI applications in the field of health management are built on the basis of group statistical laws — that is, based on large-scale population data to derive "how a certain type of crowd will perform on average". But for specific individuals, this "average law" is often difficult to apply accurately. Truly valuable health judgment requires knowing "what changes have occurred to this individual relative to their own baseline" rather than "how they compare with others".
This is exactly the core entry point of Being Labs. The company's technical route combines multimodal perception, time series prediction, individual status modeling and long-term memory, and makes comprehensive judgments by the central model through unified representation fusion. This mechanism continuously accumulates history and updates status around the same individual, identifies what changes have occurred relative to its own baseline, explains the changes in combination with environment and history, and predicts the next step. The corresponding commercial value lies in more individual-adapted judgments, more targeted reminders, and traceable explanations.
This direction is not without academic basis. Research on Google's SensorFM has shown that multimodal physiological signals can form a transferable unified representation, covering 35 health prediction tasks such as cardiovascular, metabolism, and sleep. Research involving institutions including the University of Cambridge further proposes that continuous and personalized health prediction capabilities should be developed around long-term individual trajectories, multimodal fusion and dynamic reasoning. Being Labs is building a model system from continuous observation to individual status understanding and prediction around this direction.
In terms of R&D progress, Being Labs has completed the initial development of the core model HE-VLM (Human-Centric Vision-Language Model), and formed a prototype Demo that can be demonstrated externally. At the data level, the team has collected real longitudinal individual data of hundreds of animal species, covering dozens of different modalities, including multi-dimensional information such as video, sound, motion posture, and physiological signals. These real longitudinal data provide key support for model training and individual status modeling. The team is also conducting internal benchmark tests to verify the feasibility of the method on core tasks such as individual status identification and abnormal warning.
Chief Scientist Has Served in Meta and XPeng, the Team Has Launched a Demonstrable Prototype
The team configuration of Being Labs forms complementarity in the two dimensions of technical depth and industrial experience.
Professor Li Keqian, the chief scientist, is a professor at East China Normal University, graduated from the Yao Class of Tsinghua University for his bachelor's degree, and obtained his doctorate at UCSB. He once served as a R&D lead at Meta, focusing on basic model training and user understanding; later, he served as the head of global multimodal basic models at XPeng, with practical experience in the engineering implementation of large-scale multimodal models. The project needs to solve the problems of multimodal fusion, long-term individual understanding and model training at the same time. Li Keqian's academic research and R&D experience of basic models in large factories directly correspond to these tasks, which is the core technical advantage of the team.
Yu Hongyue, the CEO, is an EMBA graduate from China Europe International Business School, once served as the first head of the pet industry at Douyin and the head of brand integrated marketing at Alibaba, with experience in platform-level industrial operation, customer cooperation and ecological expansion. The team also brings together data engineers, animal behavior experts and vertical consultants, covering the whole chain of model research, data governance, professional labeling, effect review and industrial delivery.
In terms of technical architecture, Being Labs adopts the method of integrating multiple types of expert models to continuously update individual status. Different modalities (video, sound, sensor, time series) are processed by their respective expert models, and then fused and judged by the central model. This design not only ensures the professionalism of processing each modality, but also realizes cross-modal comprehensive reasoning at the unified representation level.
Starting from Animal Scenarios, Extending to Human Health and Elderly Care
Being Labs chooses to cut into the commercialization path from the animal health scenario first, and then extend to human health.
In the animal field, the company's capabilities cover pharmacological and toxicological observation of laboratory animals, high-value livestock monitoring, pre-disease warning for aquaculture and pet health management. The company accumulates shared capabilities around long-term individual modeling, and then adapts according to species and industry tasks to expand the scope of services. The advantage of this path is that data collection in animal scenarios is relatively controllable, and commercial customers have clear willingness to pay, which can quickly verify model capabilities and form a positive cycle.
In the direction of human health, Being Labs' long-term goal is to cover the whole chain from consumer health to professional services: in the field of wearable and health management, it connects watches, rings, earphones and sleep devices to provide analysis of individual status, sleep, stress and recovery, supporting personalized health subscriptions; in the field of sports and rehabilitation, it combines movement, physiological load and recovery trajectories to serve training management, rehabilitation progress tracking and program effect evaluation; in the field of elderly care, it combines radar, camera and wearable data to track changes in activities, gait, sleep and life patterns, supporting continuous care at home and in institutions; in the field of chronic disease and off-hospital follow-up, it connects monitoring equipment, symptoms and medication records to help professional institutions grasp the status changes during the interval between visits and track intervention responses; in the field of insurance and health services, it supports long-term health management, risk stratification research and intervention effect tracking, connecting insurance, equipment and health services.
On the commercial side, Being Labs has formed a number of clearly advancing leads. In the direction of pet hardware, the first round of communication has been carried out and the second round of proposal is being prepared; in the direction of pet food, preliminary intention communication has been completed, and the first round of proposal and cooperation transformation are being promoted; in the direction of human ward management, pilot intention communication has been carried out and the second round of Demo is being promoted; in the direction of insurance companies, cooperation communication is waiting for Demo demonstration; in the direction of pet medicine and pet food brands, preliminary planning and pilot communication of AI on the production side have been carried out. In the direction of pre-disease warning for shrimp farming, the technical solution has been submitted to a city in Jiangsu, and in the direction of pre-clinical intervention cooperation for large animal toxicology, the technical improvement plan for cost reduction and efficiency increase has been submitted to a state-owned enterprise in Shanghai. The above cooperations are currently in the stages of intention, proposal and Demo, and are advancing to pilot agreements and paid verification. Among them, ward management has formed specific scenario communication, and human application has business leads for further promotion.
In terms of business model, Being Labs delivers through APIs, SDKs and end-side modules, and plans to adopt a charging structure of "joint development fee + continuous model service fee + gain/loss sharing". In the initial stage of cooperation, the effect and delivery standards are clarified through paid verification, and then continuous charging is made according to authorization, call volume or active individuals; with the increase of customers, data processing, model and deployment capabilities are gradually reused to form economies of scale.
From the perspective of industry trends, multimodal large models are moving from general conversation to in-depth applications in vertical fields. Understanding the status of individual living beings is a data-intensive, model-intensive direction that requires long-term accumulation. Once a positive cycle of data and models is formed, the barriers will continue to deepen. Being Labs chooses to cut into the animal scenario to accumulate longitudinal data and engineering experience, and then extend to the broader market of human health. The feasibility of this path remains to be verified by subsequent paid pilots and commercial data.