Building the AI decision layer for programmable immunity. The innate immune system is becoming increasingly observable and programmable in ways it never has been: we can image living immune systems, perturb them, sample them, and collect rich longitudinal data.But we still lack AI systems that can track the evolving state of an immune response and decide what should happen next. Innate responses are fast, transient, and deeply context-dependent. The moment that would be most informative can pass within hours, and the underlying state is only ever partially observed. Most models analyze experiments after the fact. The bigger opportunity is to build systems that can reason during the experiment: inferring what is uncertain, what is reproducible, and what intervention would be most informative.

The goal is to move from passive analysis to closed-loop scientific reasoning: models that infer latent biological state, critique candidate hypotheses, surface reproducibility or safety concerns, and recommend the next experiment or intervention.

Closed-loop scientific reasoning offers a path to modulating innate immunity with precision: interpreting complex immune readouts, proposing the next perturbation, and catching safety concerns early, while being grounded in real experimental platforms where imaging, manipulation, sampling, and validation are already connected.

Sophia Abraham is an AI researcher with expertise in computer vision, whose work has spanned AI safety, technical AI governance, AI systems for scientific discovery, and mechanistic interpretability. Prior to the Encode Fellowship, she worked at X, the Moonshot Factory on AI for planetary health where she developed vision systems for biological monitoring in real-world settings that later spun out. She earned her PhD in Computer Science and Engineering from the University of Notre Dame, where she focused on multi-objective optimization for robust and trustworthy AI systems. Her work has spanned AI safety, technical AI governance, AI systems for scientific discovery, mechanistic interpretability and biotech venture creation.