Medra Launches AI-Driven Platform for Continuous Scientific Experimentation
Medra, a San Francisco–based startup, has unveiled its Continuous Science Platform, a system designed to accelerate experimental research by integrating robotics with proprietary AI models. The platform is presented as a response to persistent data scarcity in scientific fields, which has constrained the development of advanced AI models compared to multimodal systems trained on vastly larger datasets.
The Continuous Science Platform operates as a closed-loop framework built from two main components:
- Physical AI automates laboratory instruments through general-purpose robots paired with agentic AI models capable of visual and language understanding. The system reportedly can operate up to 70% of standard laboratory equipment and records every movement, image, and action at high resolution, creating a structured metadata layer described as “Infra-data.” Physical AI also captures videos and metadata for each sample, detects errors with computer vision, and resumes protocols without interruption. Its modular design allows instruments and modules to be swapped easily, while agents remain instrument-agnostic, controlling both manual and automated devices.
- Scientific AI processes Infra-data together with information from electronic lab notebooks and scientific literature. These reasoning models generate experimental hypotheses and suggest follow-up actions, enabling the system to iteratively refine experimental protocols. Scientific AI supports natural language programming—allowing protocols to be adjusted in plain English, in writing or by voice—and applies multi-modal reasoning across unstructured text, images, protocols, and results. Integrated into existing notebooks, models, and workflows, it designs, executes, and adapts experiments in a closed loop. This enables researchers to create and run multiple experiments in parallel.
According to the company, this architecture is designed to compress decades of discovery into months by generating large-scale experimental datasets, which in turn train more capable scientific AI models.
Medra reports that the platform is already in use with large biotechnology and pharmaceutical companies for applications including antibody engineering, gene therapy development, and cell-based assays. Case studies released with Addition Therapeutics and Lila describe how the system was applied in real-world discovery campaigns.
With Lila Sciences, Medra’s Physical AI automated protein characterization on the LabChip GX Touch, a device considered resistant to full automation. Within three months, the system reportedly executed manual steps autonomously, scaled runs from 12 to 384 samples without hardware changes, and reduced deployment timelines from the industry average of more than a year to just 3 months. Scientists were able to modify and execute protocols independently, without requiring internal automation engineering.
At Addition Therapeutics, Medra’s closed-loop system combined Physical AI and Scientific AI to optimize RNA transfection experiments. The platform reportedly cut control variance from 36.6% to 3.7% in one month, improved transfection efficiency by up to 35%, and tripled weekly throughput, saving researchers an estimated 50 hours of manual work per week.
Cover image copyright Medra
Topic: AI in Bio
