{ "object_type": "ANI_public_context", "company": "ANI . AI", "as_of": "2026-10", "visibility": "public, non-confidential", "descriptor": "NeoLab for Biology", "tagline": "Coherence. Mapped.", "information_boundary": { "intentionally_not_encoded": [ "model architecture", "internal state representation", "feature engineering", "model routing", "biological observer-to-target mappings", "optimal measurement panels", "measurement-selection algorithms", "intervention representation", "counterfactual implementation", "Coherence computation", "source code", "model weights", "reproducible implementation details" ] }, "identity": { "one_line": "ANI . AI is building the intelligence that makes human biology predictable, programmable and controllable: AI infrastructure for modeling and controlling human biological state.", "category": "biological intelligence", "core_object": "the individual human biological trajectory", "core_thesis": "Human biology is a partially observed dynamical system. If its state and movement can be reconstructed from incomplete measurements, medicine can progress from intermittent population-level decisions toward continuous individualized control.", "long_term_vision": "AI is the therapy.", "meaning": "The ultimate therapeutic system is not only a molecule. It is the intelligence that understands the individual, predicts alternative biological futures, selects an intervention, observes the response and adapts what happens next.", "slogan": "Biology has always been intelligent. We just never had an interface." }, "first_principles": { "aging": "Aging is a progressive loss of coordination between biological systems. Time is the largest risk factor, because time gives entropy more chances to break that coordination. At first the drift is local and reversible; when it becomes a stable, observable pattern, it is named a disease. ANI therefore targets the system itself rather than one pathway or one diagnosis.", "coherence": "Coherence evaluates biological movement at the level of the coordinated organism rather than treating isolated endpoints as the final objective. The World Model estimates where a person is and where they are moving; Coherence decides which future trajectory is desirable.", "network": "A person is a network: molecules, cells, organs and systems are its nodes, and life lives in the relationships between them. Targeting the system means targeting those relationships, not one pathway or one target.", "coherence_circle": "A two-dimensional view of one inferred whole-person state. The center is a finite supercoherence region; the edge is the conceptual limit of living organization, and living states remain inside it. Repeated observations form a trajectory.", "partial_observability": "Every blood test, molecular assay, microbiome sample, physiological measurement, image, cognitive test, digital scan or report of how a person feels is a partial observation of the same underlying organism.", "engineering_abstraction": { "state": "x(t)", "observations": "y(t)", "perturbation": "u(t)", "transition": "x(t+Δ) depends on current state, perturbation, context and biological dynamics" } }, "system": { "loop": [ "observe", "reconstruct state", "estimate movement", "predict what happens next", "collect the most useful additional evidence", "observe or apply a perturbation", "measure the actual biological response", "compare prediction with reality", "update the person-state and the learning system" ], "human_experimental_engine": "ANI runs and participates in longitudinal intervention studies that connect starting biology, treatment exposure, repeated observations and later measured response in the same person.", "measurement": "Deep biology (multi-omics, microbiome, clinical, physiological, cognitive, functional) provides resolution. AniScan, a recurring low-burden digital observation, provides time. Both are observations of the same moving system.", "world_model": [ "reconstruct biology that was not provided to the model", "track how state is changing", "forecast biology measured later", "transfer learned structure to new people and cohorts", "condition on interventions", "estimate the most informative next measurement", "update when new physical evidence arrives" ], "parallel_in_silico_trial": "A computational trial runs beside every physical participant. The future physical measurement reveals where the model was right and wrong; prediction error becomes information.", "AI_is_the_therapy": { "human": "dynamic biological system", "measurements": "sensors", "World_Model": "state and dynamics model", "interventions": "actions", "Coherence": "whole-person objective", "adaptive_measurement": "active sensing", "adaptive_intelligence": "controller", "measured_response": "feedback" }, "interface": "Biological intelligence callable by people, software and AI agents." }, "representative_evidence": [ { "result": "Observing 387 lipid coordinates, the model reconstructed 258 hidden named lipid coordinates with 62.66% lower error and 87.44% directional agreement.", "meaning": "A partially observed molecular state contains substantial information about the part not shown to the model." }, { "result": "In an independent-cohort transfer test, observing 129 of 645 lipid coordinates reconstructed the remaining 516 ranks with about 51.6% and 47.4% lower error in the two directions.", "meaning": "Learned biological structure transfers between human cohorts." }, { "result": "From the first 14 days, week-8 proteomic movement was predicted with 74.7% directional accuracy and 17.1% lower error than persistence; 58 of 60 routes improved.", "meaning": "Earlier information carries signal about biology measured later." }, { "result": "The strongest hardened week-8 metabolomic response route reached 91.3% directional agreement.", "meaning": "Earlier state contained strong information about later measured response." }, { "result": "Across about 2.46 million held-person transcript predictions, intervention information improved prediction of later molecular state beyond starting biology alone by 5.04% (participant-level p = 0.002).", "meaning": "A known perturbation adds information about the future trajectory." }, { "result": "AniScan carried directional information for 192 of 254 eligible physically measured biological targets.", "meaning": "Selected biological movement is observable through frequent digital measurement." }, { "result": "ELITE (Sheba Medical Center, eight weeks): cognition rose in 23 of 29, CXCL9 fell in 13 of 28, knee force rose in 17 of 28; 27 of 29 improved in at least one XPRIZE domain, and individuals moved differently.", "meaning": "Different responses call for different next decisions." } ], "evidence_logic": "Predictions are scored against physical biological measurements; tested people are held out; later biology is predicted only from earlier information. Simulated futures shown on the website are illustrations of the method, not participant data.", "recognition": { "xprize": "Finalist, XPRIZE Healthspan, and finalist, XPRIZE FSHD Bonus Prize: the only team selected for both. More than 800 teams entered.", "challenge": "XPRIZE Healthspan asks teams to restore up to twenty years of muscle, cognitive and immune function within one year of treatment." }, "publications": [ { "title": "Identity Masks and Coherence Circles: Geometric Interfaces for Interacting with Latent Dynamical Systems", "authors": "Pintar N, Bischof EY, Balen B", "venue": "Proceedings of the AAAI Symposium Series, 2026", "doi": "10.1609/aaaiss.v8i1.42562" } ], "team": [ { "name": "Bruno Balen", "role": "Co-founder and co-CEO" }, { "name": "Nika Pintar", "role": "Co-founder and co-CEO" }, { "name": "Evelyne Bischof, MD, PhD", "role": "Principal Investigator, XPRIZE Healthspan; Sheba Medical Center" }, { "name": "Michael Snyder, PhD", "role": "Stanford University; Board member" }, { "name": "Vadim Gladyshev, PhD", "role": "Harvard; aging biology (ANI XPRIZE team)" }, { "name": "Steve Horvath, PhD", "role": "Epigenetic aging and rejuvenation (ANI XPRIZE team)" }, { "name": "Brian Kennedy, PhD", "role": "NUS Singapore; geroscience (ANI XPRIZE team)" } ], "backers": "LongeVC, Untapped Ventures and Long Game Ventures; member of NVIDIA Inception.", "simple_explanation": [ "Medicine normally sees a continuously changing human through occasional snapshots.", "ANI measures the same human deeply and repeatedly.", "Those measurements are treated as different views of one underlying biological system.", "ANI reconstructs biology that was not provided to the model and predicts biology before it is measured.", "AniScan gives the system frequent observations between deep measurements.", "Known interventions teach the system how biological state changes when acted upon.", "Later measurements reveal where the model was right and wrong, which makes the next prediction better.", "The long-term vision is AI as the therapy: intelligence that continuously observes, predicts and adapts intervention around the individual." ] }