proc.bio
Version-controlled bioprocesses with agentic review and a marketplace
Software engineering solved the knowledge-sharing problem decades ago with version control, pull requests, and structured peer review. Biomanufacturing has no equivalent - protocols remain locked in static SOPs, proprietary systems, and spreadsheets, rarely shared in formats that are machine-readable, composable, or amenable to computer-aided review.`
We are creating infrastructure for collaborative development across the scales of molecular product design and process engineering - a "GitHub for bioprocesses" - that pairs version-controlled collaboration with automated, multi-tier hybrid (physics- and inference- based) agentic review. Specialized AI agents evaluate protocol contributions across three levels of abstraction: molecular and reaction feasibility agents assess thermodynamic plausibility and biocatalytic compatibility; process performance agents evaluate kinetic parameters and scale-up predictions; and techno-economic (TEA) and process hazard (PHA) agents score cost-of-goods projections and safety risk profiles. Feedback loops allow downstream results to trigger upstream re-evaluation, shifting feasibility, economic, and hazard review from periodic manual exercises to continuous, embedded quality gates. Users can also compose custom automation workflows, chaining agents, validation checks, and notifications to match their own development processes.
Asset piles - from molecular designs to process flow diagrams - are encoded in standardized formats building on emerging standards such as LabOP. To lower the barrier to adoption, a protocol translation agent assists in ingesting natural-language SOPs into structured, ontology-grounded representations. Repositories support both open-source and private access, enabling organizations to share pre-competitive knowledge publicly while maintaining secure environments for proprietary IP. Research groups can contribute new models and agents, commercial data providers can return manufacturing data for model training, and national infrastructure networks provide pilot facilities and data pipelines that close the design-build-test-learn loop at scale.