Skip to content

4ailabs/inferentia

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Inferentia

Built with Opus 4.7 · Cerebral Valley × Anthropic Hackathon · April 2026


Core thesis

An ill person is an organism executing predictions across its four cognitive levels — cellular bioelectric, organic homeostasis, somatic unconscious, explicit consciousness. Altered biomarkers, chronic symptoms, metabolic rigidity: all are the signature of instructions the organism keeps executing, not memories it consults. Inferentia identifies those active instructions — predictive imprints, deficiencies limiting regulation, toxins forcing the system, available agency, amplifying genetics — and computes which interventions expand the frontier the pattern contracted. The outcome is not normalizing a value: it is enabling the organism to do more.

Tesis (original, español):

Una persona enferma es un organismo ejecutando predicciones en sus cuatro niveles cognitivos — bioeléctrico celular, homeostasis orgánica, inconsciente somático y consciencia explícita. Biomarcadores alterados, síntomas crónicos, rigidez metabólica: todo es la signatura de instrucciones que el organismo sigue ejecutando, no recuerdos que consulta. Inferentia identifica esas instrucciones activas — improntas predictivas, carencias que limitan la regulación, toxinas que fuerzan el sistema, agencia disponible, genética amplificadora — y calcula qué intervenciones expanden la frontera que el patrón contrajo. El outcome no es normalizar un valor: es que el organismo pueda más.

What Inferentia does

Inferentia is a clinical tool that identifies active predictive instructions across six co-present layers — active imprints, nutritional substrate, toxic load, available agency, inferred nutrigenomic profile, and observable signature — and computes interventions that expand the organism's regulatory capacity. The computational engine combines Active Inference (Friston) with multivariate Bayesian classification over clinical signatures, integrating the TAME multi-scale cognition framework (Levin), phenotypic flexibility (van Ommen), and the BV4 clinical framework (Ojeda Rios, proprietary).

Status

Research prototype — built during Cerebral Valley × Anthropic Hackathon (2026-04-21 through 2026-04-26).

  • Not a medical device.
  • Not a diagnostic tool.
  • Not a replacement for professional medical consultation.
  • Clinical validation: planned for Phase 2, post-hackathon.

Architecture (two-level, clinician-validated)

Patient view     ←→  Unified backend engine  ←→     Clinician view
(narrative        Opus 4.7 orchestrator         (technical posteriors,
accessible        + Active Inference (pymdp)    SNPs, metabolic
predictions,      + Bayesian inference          signature, clinical
agency panel)     (NumPyro)                     note, editable plan)
                  + Multimodal fusion

All clinical prescriptions require clinician validation before reaching the patient. The patient sees their own predictive map and agency metrics; the clinician sees the full technical stack and validates.

Five MVP outputs

  1. Predictive Map — hero visualization of active predictive priors.
  2. Expected Metabolic Signature — inferred biomarkers with discordance vs. labs.
  3. Prioritized Nutrigenomic SNPs — Bayesian posterior over relevant SNPs (clinician view only).
  4. Salutogenic Nutritional Program — clinician-validated, agency-oriented.
  5. Agency Panel — pre/post delta on primary outcome (not biomarker — agency).

Theoretical framework

Pilar references (not exhaustive):

  • Friston (2010, 2017) — Free Energy Principle / Active Inference.
  • Sterling & Eyer (1988); McEwen (1998, 2017) — Allostasis / allostatic load.
  • Barrett (2017) — Predictive interoceptive construction.
  • Porges (2011) — Polyvagal Theory.
  • Levin (2019) — TAME framework / cognitive cone.

The clinical instance used in this prototype references the author's independent clinical framework (BV4 Clinical Treatise, Dr. Miguel Ojeda Rios) — a taxonomy of 13 survival imprints with canonical Spanish names. That framework is proprietary to the author and is not included in this repository. This hackathon project is a distinct new implementation that references BV4 only as one of several possible clinical ontologies mappable to the Active Inference substrate. For a one-page conceptual overview including Spanish-English term equivalences, see docs/concepts/imprints_overview.md.

Originality and scope declaration (hackathon compliance)

  • All code in this repository was written during the hackathon (2026-04-21 onward).
  • No prior code, models, or datasets have been reused.
  • No real patient data is used anywhere. All demonstrations use a synthetic patient generated by the author.
  • The author's independent clinical framework (BV4) is referenced as intellectual context but is not included in code or data form in this repo.

Open source and licensing

  • This project is released under the MIT License (see LICENSE).
  • All dependencies are open source. See CREDITS.md for the full list and respective licenses.
  • Public biomedical datasets used (see CREDITS.md): GWAS Catalog (EBI, CC0), gnomAD (CC0), Human Phenotype Ontology (CC-BY), Reactome (CC0).
  • Psychometric instruments: Pearlin Mastery Scale (public domain, Pearlin & Schooler 1978). Interoceptive items are author-derived; no copyrighted instrument has been copied.

Usage policy compliance

This project operates within Anthropic's Usage Policy. Specifically:

  • The system does not provide autonomous medical diagnosis or treatment recommendations to end users.
  • All clinical outputs are presented as probabilistic hypotheses requiring professional validation.
  • The two-level architecture (patient view + clinician validation gate) ensures human oversight of all clinical content delivered to patients.
  • Language in the patient-facing interface is strictly probabilistic and non-directive.

Disclaimers

This prototype is for research and demonstration purposes only.

  • Not approved by any regulatory body (FDA, EMA, COFEPRIS, etc.).
  • Not intended for clinical decision-making in production settings.
  • Not to be used as a substitute for professional medical advice, diagnosis, or treatment.
  • The author assumes no liability for use outside the controlled demonstration context of the hackathon.

Team

  • Dr. Miguel Ojeda Rios — Author, clinical lead, full stack (solo team).

Submission

  • Hackathon: Built with Opus 4.7 (Cerebral Valley × Anthropic).
  • Problem statement: #1 — Build From What You Know.
  • Submission deadline: 2026-04-26, 20:00 EST.

About

Medicina Clínica de Inferencia Activa. Built with Opus 4.7 — Cerebral Valley × Anthropic Hackathon, April 2026.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors