VISTA Platform

The VISTA Platform, developed by JURA Bio, Inc., is an AI-driven infrastructure designed to overcome the fundamental data bottlenecks that constrain frontier model training in therapeutic discovery. By integrating AI directly with DNA synthesis chemistry and high-throughput experimental workflows, VISTA enables continuous self-improvement of foundation models at a scale that conventional automated laboratory systems cannot achieve. The platform is built for organizations working on next-generation therapeutic design challenges, including antibody and cell therapy development, regulatory promoter engineering, and emerging biological modalities.

At the core of JURA's approach is the recognition that public databases lack the functional annotations — binding specificity, off-target activity, and developability — that matter most for training capable AI models. Conventional lab workflows cannot generate experimental data fast enough to meaningfully advance frontier models, particularly for the hardest targets and newest modalities. VISTA addresses this by creating closed-loop, AI-controlled data generation cycles capable of producing over 1 billion effective laboratory observations per training loop, compounding into proprietary functional datasets that continuously improve candidate quality.

Core Platform Components

• Variational Synthesis: AI-controlled DNA synthesis that designs and directs stochastic chemical reactions to efficiently sample from generative models of biological sequence. This approach replaces the conventional GPU → CUDA → LLM stack with direct on-chip sampling and synthesis, enabling physical construction and functional screening of libraries spanning over 10,000 trillion (10^16) AI-designed candidate sequences in vitro.
• LIFT: AI-interpretable experimental conditions that scale data capture. LIFT translates high-throughput wet-lab measurements into training signal optimized for model training, enabling massive parallel data generation per campaign and delivering up to 100x data throughput compared to standard approaches.
• LeaVS: Co-designed inference systems built to train sovereign models as rapidly as possible. LeaVS maximizes the information gained per experiment, ensuring each wet-lab cycle yields the highest possible acceleration in model improvement — achieving up to 100x learning acceleration through codesigned inference.

Platform-Scale Capabilities

• Screening of over 10,000 trillion AI-designed therapeutic candidate sequences in vitro within a single platform
• 100x single-cell readout capacity in AI-interpretable experiments
• Data training loops incorporating 1 billion or more effective laboratory observations
• Continuous feedback from each experimental cycle into JURA's foundation models, compounding proprietary functional data over time
• Support for multiple therapeutic modalities including antibodies, cell therapies, and regulatory promoters

Applications and Use Cases

• VISTA in action: AI-controlled data generation loops in vitro, demonstrating closed-loop model training at unprecedented scale
• MESA: Multiplexed biologics discovery at scale, showcasing the platform's application to high-throughput candidate identification
• Optimal experimental design: A rethinking of experimental design principles to maximize information gain per wet-lab cycle for AI model training, as outlined in JURA's perspective on running one billion simultaneous experiments

JURA Bio's VISTA Platform represents a vertically integrated approach to AI-driven therapeutic discovery, combining proprietary synthesis chemistry, interpretable experimental infrastructure, and co-designed inference systems into a unified loop. Each component is purpose-built to ensure that frontier models can self-improve continuously, generating the functional biological data necessary to tackle the most challenging targets in modern drug development.