July 2-3, 2026, the 2026 AI & Biopharma Ecosystem Conference opened in Shanghai. OxTium Technology made a major appearance with its BioFord Agent embodied intelligence discovery platform and industrial-grade fermentation closed-loop solutions, attracting researchers and pharma representatives to its booth, making it one of the highlights of the event.
At the conference, OxTium Technology's Co-founder & CTO Deng Siwei delivered a keynote speech titled "AI Agent-Driven Full Lifecycle of Biopharma: From Laboratory Agents to Industrial Closed-Loop Control."
Deng pointed out that the core bottleneck in the biopharma industry is never "AI not being smart enough," but rather severe process fragmentation, difficulty in transferring experience, and each stage involving repetitive labor and "black-box" steps that rely heavily on human expertise. The essence of the problem is the disconnect between cognitive and execution capabilities—large models can think but cannot act; lab equipment can be programmed but lacks intelligent scheduling.
"Let AI truly run into labs and factories, transforming the cognitive capabilities of large models into reproducible productivity."
Laboratory Side: Physical AI Breaks Down Equipment Protocol Barriers
Deng believes that labs have long faced a seemingly simple yet extremely challenging issue—chaotic equipment scheduling. Multiple instruments, research groups, and projects with different priorities intertwine, making equipment collisions and rescheduling a daily occurrence. Moreover, different instrument models have varying protocols, making manual coordination costly.
OxTium's experimental scheduling agent uses a Universal Instrument Abstraction Layer to break down protocol barriers across heterogeneous devices. Whether it's PCR machines, microplate readers, flow cytometers, or automated liquid handling workstations, all can be uniformly managed and scheduled. The system's built-in dynamic scheduling algorithm automatically batches schedules, avoids conflicts in real-time, and automatically records all process parameters, forming a traceable experimental audit trail.
This means AI is no longer just a "paper strategist" that produces plans, but a true "research collaborator" that enters the lab, operates equipment, and executes tasks.
In addition, BioFord Agent integrates a literature retrieval agent (BioScholar), an experimental design agent (BioDesigner), a scientific agent (BioModeler), and a data analysis agent (BioAnalyst). These five agents share a unified knowledge base and collaborate to complete the full research loop from literature review to result reporting.
Industrial Side: AI-Driven Fermentation Process Optimization
On the industrial side, Deng demonstrated how OxTium extends the same "sense-predict-decide" AI engine seamlessly from the lab to the factory floor.
Taking biological fermentation as an example, the industry has long faced four major pain points: invisible strain status, experience-based feeding, late anomaly detection, and difficult data accumulation, leading to high batch titer variability. OxTium's solution aggregates multi-source data from DCS, sensors, and offline testing through a digital cockpit, combined with time-series models to achieve soft sensing (online estimation of cell density/titer), automated feeding, anomaly alerts, and endpoint prediction, directly linking to the factory's DCS system to issue control commands.
AI compresses batch titer variability into a stable range. In one peptide project, it achieved an 18.4% reduction in unit production cost, a 10% reduction in byproduct concentration, and real-time monitoring latency reduced from 2 hours to under 1 minute.
Deng emphasized that the difference between labs and factories is not about "whether to use AI," but rather the focus on data formats, control priorities, and deployment strategies. OxTium adopts a three-tier private deployment architecture: cloud central layer + factory edge layer + workshop on-site layer, ensuring both model cognitive depth and meeting the stringent real-time and security requirements of industrial scenarios.
From lab to factory, OxTium is reshaping the R&D ecosystem with "Physical AI": enabling models to act, data to flow back, and experience to be transferred. In the future, we will continue to deepen our investment in life science AI infrastructure, making every scientific discovery truly reproducible productivity.