Zhangtongshe zhangtongshe.com
Building Foundation Models for Life Sciences.
Recently, Shanghai Guanyi Intelligent Technology Co., Ltd. (hereinafter referred to as "Guanyi Intelligence") announced the completion of its first funding round totaling hundreds of millions of CNY. The round was led by IDG Capital, with participation from CCV Partners, Yunbai Capital, Jiaoda Hanyuan Assets, and Yiqi Capital. The funds will primarily be used for R&D on life intelligence foundation models, construction of wet lab platforms, and industrial collaboration validation.
Founded in November 2025 and registered in Xuhui District, Shanghai, Guanyi Intelligence emerged from Shanghai Jiao Tong University (SJTU). Amid the wave of AI for Science, this company has chosen a highly ambitious yet challenging path—building foundation models for life sciences.
From Computer Vision to Life Sciences
Guanyi Intelligence was initiated by Professor Wang Yanfeng from the School of Artificial Intelligence at SJTU. Zhang Ya and Xie Weidi are co-founders, with Xie Weidi serving as CEO and Chief Scientist.
Xie Weidi's research journey did not begin in life sciences. After studying communication engineering as an undergraduate, he went to University College London in 2013 for his master's degree, focusing on using computer vision to recognize handwritten Arabic text.
At that time, the potential of deep learning was not widely recognized. He self-studied relevant methods through public courses to complete his research project, then joined the Visual Geometry Group (VGG) at Oxford University for his PhD under Andrew Zisserman and Alison Noble, receiving the inaugural Oxford-Google DeepMind scholarship. After graduation, he remained at Oxford for postdoctoral research before returning to China in 2022 to join SJTU.
After years of dedication to the visual domain, he keenly observed that core algorithms in traditional computer vision had become relatively mature, with future focus shifting towards scaling up. A grander idea began to take shape: given the powerful capabilities of large language models and vision models, could they be used to solve specific scientific problems? Leveraging his prior experience in medical imaging, the integration of AI and life sciences became a natural entry point.
In the summer of 2025, Xie Weidi, along with co-founders Wang Yanfeng and Zhang Ya, began preparing to establish the company. Zhang Ya is currently Vice Dean and Distinguished Professor at the School of Artificial Intelligence, SJTU, possessing a background in both biology and artificial intelligence, with long-term engagement in interdisciplinary research on multimodal AI and healthcare.
The decisive factors were the qualitative leap in AI model capabilities and the reshaping of the industrial landscape.
Through their research in life science language models, the team realized that large language models trained independently by academia could no longer match those from industry. Large models have entered a phase where they are broadly "usable." Academia no longer has the resources to compete in general-purpose large models but can leverage existing pre-trained models for deep integration in vertical domains. Meanwhile, the cost of collecting multimodal data is decreasing, while exploring the upper limits of model capabilities requires sustained capital investment.
Guanyi Intelligence chose to first borrow the capabilities of general models and then train deeply for life sciences. The company does not plan to train general-purpose large language models from scratch on a massive scale. Instead, it utilizes existing pre-trained models to conduct subsequent training and multimodal integration oriented towards life sciences, validating effectiveness in specific fields first and iterating continuously.
Building the "Simulation Base" for Life Sciences
In the broad landscape of AI for Science, Guanyi Intelligence has not chosen to build single-point scientific tools but aims to become a foundation model company for life sciences.
General AI excels in logical reasoning and language processing, but life sciences require a true understanding of biological system dynamics. Traditional biological research is often limited to single levels, whereas Guanyi Intelligence attempts to bridge cross-level relationships from molecules and cells to tissues and disease phenotypes.
Guanyi Intelligence will integrate biological information across molecular, cellular, and tissue levels, along with disease phenotypes and related medical knowledge, to learn the connections between different life levels. On this basis, the company will construct a life simulation model centered on virtual cells, studying cell states and functional responses under interventions such as genetic changes and drug actions, and validating model predictions through real experiments.
The versatility of this foundation model lies in the ability to accumulate and reuse capabilities across tasks: understanding cell states supports disease mechanism research; predicting intervention responses helps evaluate potential targets and candidate drugs; learning from experimental results further supports the design of the next round of experiments.
The company will gradually test the model's transfer capability across these interconnected tasks in various research scenarios, with the long-term goal of exploring batched drug discovery.
Currently, the team has achieved substantial validation in several frontier directions.
In the field of rare disease diagnosis, the DeepRare system developed by the team launched in-hospital quality control applications at Shanghai Xinhua Hospital in February 2026. The system can automatically analyze outpatient cases and flag suspected patients.
It is reported that DeepRare has completed screening for over 1.1 million outpatient visits, detecting more than 2,500 suspected rare disease cases, improving the identification rate of suspected rare diseases by approximately 40% compared to conventional processes.
Additionally, the team's research on tumor pathology multimodal understanding (KEEP) and virtual cell drug perturbation prediction (MAP) has been published in top journals such as Cancer Cell and Nature Machine Intelligence, validating its capabilities in evidence-based reasoning, multimodal understanding, and unknown intervention prediction.
Guanyi Intelligence is continuously expanding life science data, knowledge, and modalities, building a high-quality perturbation data engine, and establishing a loop of "model prediction—experimental validation—data feedback—model update."
Guanyi Intelligence plans to build its own wet lab platform, integrating model R&D, biological observation, and experimental validation into a single R&D closed loop. The model integrates existing knowledge and experimental data to propose mechanistic hypotheses and assist in experiment design; researchers validate predictions through experiments, feeding new evidence back to the model to support the next round of learning and research, driving AI from learning static data to actively exploring the living world.
In terms of commercialization, Guanyi Intelligence plans to offer callable bio-intelligence capabilities via Model-as-a-Service to research institutions, hospitals, pharmaceutical companies, and biotechnology firms, serving scenarios such as disease research, drug discovery, protein design, and synthetic biology.
A Track Betted On by Giants
Life sciences are becoming a key strategic direction for leading global AI enterprises.
OpenAI has launched GPT-Rosalind for life science research, combining biological reasoning with scientific tools and research workflows. Anthropic has also introduced Claude Science, deeply engaging in life science experimental research and drug discovery collaborations.
AI is moving beyond simple text generation to deeply participating in the entire research process, from posing scientific questions and analyzing data to experimental validation.
This is precisely the logic behind this funding round.
IDG Capital has long-standing investments in hard tech and life sciences. CCV Partners and Yiqi Capital have deep expertise in AI and frontier technologies. Yunbai Capital possesses rich resources for industrial synergy. The entry of Jiaoda Hanyuan Assets carries the endorsement of the SJTU ecosystem for this team.
The collective action of multiple institutions not only recognizes the team's combined academic and engineering background but also indicates that amidst the intensifying competition in general large models, capital is shifting its attention to vertical-domain scientific foundation models.
The appeal of Guanyi Intelligence lies not only in standing at the intersection of two major trends but also in the team's composite DNA of top-tier academic background and engineering implementation capabilities, having already validated a closed loop in real clinical scenarios. On this long and snow-covered track, this SJTU-affiliated team has just begun.
Text | Zhouyu Editor | Yida