Reference corpus
Research memoirs, deployment frameworks and doctrinal articles. Most of the corpus is freely accessible upon registration — no sales outreach.
The Twingital Institute's publications document the concrete transition from performant AI model to industrially deployable system. They are intended for decision-makers, system architects, compliance officers and general management of organizations deploying AI in critical environments.
The corpus is organized in three categories: research memoirs and reports (long, foundational documents), reference frameworks and deployment guides (operational tools), articles and doctrinal positions (position papers on open questions).
Research memoir · Twingital Institute
Medical vigilance and chronicity: systemic diagnosis and architectural properties of a sustainable regime. 2-page cabinet synthesis.
Research memoir · Twingital Institute
On what we see in consultation, what we cannot see, and what could change. A PREDICARE note for care actors.
Research memoir · Twingital Institute
Doctrine of a territorial predictive vigilance infrastructure under advanced chronicity: why prevention remains structurally non-financeable.
Research memoir · Twingital Institute
Systemic diagnosis of the French healthcare system and architectural properties of a sustainable vigilance regime under chronicity.
Research memoir · Twingital Institute
Memoir documenting the full arc of the PREDICARE programme: epidemiological foundations, architecture, validation, governance.
Reference framework · Twingital Institute
Technical note distinguishing three architectural families (world models, JEPA, memory models) as responses to the three blind spots of the LLM paradigm.
Reference framework · Twingital Institute
The FDA PCCP is the mature form of a contractualised promotion port, distinct from the model registry. Tripartite apparatus and ex ante admission regime.
Reference framework · Twingital Institute
Structural taxonomy of agentic refusals in 6 categories. Refusal as architectural primitive of governance, not as a residue of the system.
Reference framework · Twingital Institute
Apparent vs structural exogeneity. Three-layer validation architecture for AI systems in regulated contexts.
Reference framework · Twingital Institute
Hexagonal architecture as a condition of governability for regulated AI. Reliability ports, domain-infrastructure separation, normative testability.
Reference framework · Twingital Institute
Measured performance vs operational reliability: calibration, applicability domain, decisional validity — three properties regulated AI can no longer conflate.
Reference framework · Twingital Institute
Procedural vs architectural AI governance. Native auditability, validity domain, decision/automation separation.
Reference framework · Twingital Institute
Composite agentic architecture: algorithmic swarms, domain memory, and event-driven orchestration for regulated digital twins.
Reference framework · Twingital Institute
Why agentic AI is fundamentally an enterprise architecture problem. Five structural functions of EDA for situated, auditable, governable AI agents.
Article — Position paper · Twingital Institute
A model does not become sovereign because it is open. Three chains of proof: inspectable weights, documentable training data, governable versions.
Article — Position paper · Twingital Institute
Of four digital sovereignty layers, only the energy layer is allocated, not composed. AI enters macro-industrial governance.
Article — Position paper · Twingital Institute
Seven layers, three verdicts: arbitration doctrine of composite sovereignty for AI in regulated European environments. Silicon changes political nature.
Research memoir · Twingital Institute
Medical vigilance and chronicity: systemic diagnosis and architectural properties of a sustainable regime. 2-page cabinet synthesis.
Article — Position paper · Twingital Institute
Architectural doctrine of the sovereignty port: compliance computed by the gate, proven by the ledger. Three planes, three threats, AI Act, SecNumCloud.
Research memoir · Twingital Institute
On what we see in consultation, what we cannot see, and what could change. A PREDICARE note for care actors.
Research memoir · Twingital Institute
Doctrine of a territorial predictive vigilance infrastructure under advanced chronicity: why prevention remains structurally non-financeable.
Research memoir · Twingital Institute
Systemic diagnosis of the French healthcare system and architectural properties of a sustainable vigilance regime under chronicity.
Article — Position paper · Twingital Institute
CTO doctrine on citizen agentic AI, unassignable decisional debt, and the four COMEX decisions that turn unassignable risk into accountable risk.
Article — Position paper · Twingital Institute
AI kilowatt-hour allocation is triply distributed (market, contracts, local blockages) without deliberative procedure. Conditions for a sectoral protocol.
Article — Position paper · Twingital Institute
Doctrinal note distinguishing three reliability ports (disciplined split, explicit calibration, signed AD) as architectural alternative to the benchmark.
Exploration · Twingital Institute
The CVE-2026-31431 (Copy Fail) case on WSL2 kernel 6.6.87.2 as empirical terrain for the promotion gate between technical artifact and normative dependency.
Article — Position paper · Twingital Institute
The EU is not softening medical AI regulation: it is recording a categorial MDR/AI Act incompatibility it cannot resolve, externalizing it to manufacturers.
Reference framework · Twingital Institute
Technical note distinguishing three architectural families (world models, JEPA, memory models) as responses to the three blind spots of the LLM paradigm.
Exploration · Twingital Institute
JEPA and latent predictive architectures: geometry of predictability, anti-collapse constraints, the unmet threshold of biographical memory.
Article — Position paper · Twingital Institute
Regulation by friction-of-use in U.S. health insurance. Unobservability as a political variable. The concept of administrative abandonment.
Reference framework · Twingital Institute
The FDA PCCP is the mature form of a contractualised promotion port, distinct from the model registry. Tripartite apparatus and ex ante admission regime.
Reference framework · Twingital Institute
Structural taxonomy of agentic refusals in 6 categories. Refusal as architectural primitive of governance, not as a residue of the system.
Reference framework · Twingital Institute
Apparent vs structural exogeneity. Three-layer validation architecture for AI systems in regulated contexts.
Article — Position paper · Twingital Institute
Exogenous agentic governance: action-space/autonomy/reversibility risk grid, decision regimes and admission contracts.
Reference framework · Twingital Institute
Hexagonal architecture as a condition of governability for regulated AI. Reliability ports, domain-infrastructure separation, normative testability.
Reference framework · Twingital Institute
Measured performance vs operational reliability: calibration, applicability domain, decisional validity — three properties regulated AI can no longer conflate.
Exploration · Twingital Institute
ToxTwin V2.4: endpoint tri-router, enriched Ames ISS and hERG ChEMBL corpora. Tox21 mean AUC 0.898, Ames 0.853, hERG 0.800. Lessons from a controlled cycle.
Reference framework · Twingital Institute
Procedural vs architectural AI governance. Native auditability, validity domain, decision/automation separation.
Reference framework · Twingital Institute
Composite agentic architecture: algorithmic swarms, domain memory, and event-driven orchestration for regulated digital twins.
Reference framework · Twingital Institute
Why agentic AI is fundamentally an enterprise architecture problem. Five structural functions of EDA for situated, auditable, governable AI agents.
Article — Position paper · Twingital Institute
Gap between massive deployment and real adoption of Microsoft 365 Copilot. 3.3% penetration, negative NPS, generative shadow IT.
Exploration · Twingital Institute
Dreyfus, Searle, and the language/representation confusion. Explicit, implicit, and pipeline encoding. The data scientist layer.
Exploration · Twingital Institute
Biological transduction vs mediated encoding. Symbol grounding problem. Molyneux's problem, Mary's argument. Three grounding levels.
Exploration · Twingital Institute
Memory as polycentric graph. Biographical edge vs statistical edge. World models: decisive advance but insufficient threshold.
Exploration · Twingital Institute
Full article (28 pp.). Encoding, transduction, stratified grounding, polycentric memory graph, biographical edge, world models.
Article — Position paper · Twingital Institute
Why most Bayesian implementations in health AI are frequentism dressed up as a probabilistic graph. Calibration, drift, and the HR argument.
Article — Position paper · Twingital Institute
AI explainability in medicine is a relational predicate, not an absolute. Measured, audited, monitored reliability over illusory transparency.
Article — Position paper · Twingital Institute
Analysis of AI's physical constraints: energy, water, data, silicon. $200B in GPUs, <10% adoption, Jevons rebound effect.
Article — Position paper · Twingital Institute
Six-dimension eligibility framework: decision topology, time balance, uncertainty reduction, coupling, TCO, reversibility.
Article — Position paper · Twingital Institute
Structural analysis of AI-induced clinical cognitive load: techno-solutionism, collective decision-making, strained IT departments, TweenMe encapsulation.
Article — Position paper · LinkedIn
How to scale a predictive prevention infrastructure without betraying territorial equity, clinical safety and the human purpose of care.
Article — Position paper · LinkedIn
Overview of the PREDICARE series: from the systemic diagnosis of medical abandonment to a territorial predictive prevention infrastructure.
Article — Position paper · LinkedIn Pulse
From experimentation to mass production without manufacturing risk. Alert fatigue, verified interoperability, change management, clinical safety.
Article — Position paper · LinkedIn Pulse
Five levers to unlock funding: FIR/Article 51, ONDAM allocation, disciplined PPP, SIB/OIS, incentive reform. Territorial digital twins for accountability.
Article — Position paper · LinkedIn Pulse
Predictive prevention as 21st-century infrastructure. Five conditions: governance, financing, training, interoperability, outcomes-based management.
Article — Position paper · LinkedIn Pulse
The value creation/capture decoupling, the rational budgetary suicide of hospitals under fee-for-service, the BFR valley of death. Systemic analysis.
Article — Position paper · LinkedIn Pulse
Clinical scenario of André's digital twin: detection at D-21, alert at D-13, intervention at D-12. RPM evidence base and the PREDICARE project.
Article — Position paper · LinkedIn Pulse
Why the system intervenes too late by design. 5-layer architecture: IoMT, companion app, digital twins, alerts, professional dashboard.
Article — Position paper · LinkedIn Pulse
Four systemic mechanisms turn the abandonment of polypathological patients into a statistical phenomenon: demography, economics, cognition, fragmentation.
Article — Position paper · LinkedIn Pulse
12 million patients with chronic conditions, a collapsing medical workforce. How the French system produces invisible patients.
Article — Position paper · LinkedIn
The MIT 95% AI failure rate reflects measurement bias, not technology failure. Portfolio ROI, platform economics, and the Data+AI foundation.
Article — Position paper · Doctrinal position · LinkedIn
How generative AI can correct representativeness biases in clinical cohorts through synthetic data augmentation.
Exploration · Twingital Institute
GVI/PCI analysis (MIT CPO) of the French sovereign cloud market: hyperscalers, SecNumCloud, territorialised hybridisation, risk matrix.
Article — Position paper · Twingital Institute
A framework combining adaptive trials, synthetic controls and RCTs under unified governance for AI clinical evidence.
Article — Position paper · Doctrinal position · LinkedIn
Impressive technical progress does not exempt from economic and clinical proof. The missing independent health-economic evidence for AI in healthcare.
Article — Position paper · Twingital Institute
Transposing the PINNs principle to clinical medicine. Published biomedical literature integrated as a constraint in the loss function.
Article — Position paper · Twingital Institute · January 2025
AI as research accelerator in bio-industries. Data quality, HDLSS, cybersecurity, AlphaFold v3 limitations.
Article — Position paper · Twingital Institute
Why the patient digital twin concept lacks defensible scientific content — and why the answer is a generator.
Article — Position paper · Twingital Institute
From buzzword to operational discipline. Analytical framework, maturity models, ontological intelligence, cybersecurity and TweenMe.
Article — Position paper · Twingital Institute
Critical analysis of agentic AI frameworks. Historical roots, structural novelties, persistent challenges, and the cybersecurity blind spot.
Exploration · LinkedIn · January 2024
What emergency medicine and philosophy of mind say about consciousness — and its implications for AI.
Exploration · Twingital Institute
Perceptrons, transformers, emergent complexity through massification. The brain is fractal, not a lasagna dish.
Exploration · LinkedIn · April 2020
COVID-19 excess mortality in nursing homes: geriatric frailty, polypharmacy, ICU limits, palliative care ethics.
Exploration · LinkedIn · March 2020
Physiopathology of SARS-CoV-2: ACE2 binding, TMPRSS2 activation, cytokine storm, ARDS, sepsis, and therapeutic approaches (March 2020).
Exploration · LinkedIn · March 2020
RT-PCR mechanics, ELISA immunological tests, sensitivity/specificity trade-offs, and why massive testing alone cannot contain a pandemic.
Exploration · LinkedIn · March 2020
Fundamental virology refresher: viral structure, envelope, genetic payload, replication cycle, mutations and recombination of SARS-CoV-2.
Article — Position paper · LinkedIn Pulse
Technological progressivism and the refusal of death as structural inflation drivers. The SOVALDI case. Alzheimer's, cancers, metabolic syndrome.
Article — Position paper · LinkedIn Pulse
Health as a superior good per Kenneth Arrow. Bismarck vs Beveridge comparison. The American model: 18% of GDP.
Article — Position paper · LinkedIn Pulse
Mapping the French healthcare ecosystem. Institutions, agencies, care providers, payers, industry. ONDAM, T2A, PMSI.
Article — Position paper · LinkedIn
Applying NIST cloud patterns to healthcare. Standardized service catalogue, IoT self-service, elastic capacity, anywhere access.
Article — Position paper · LinkedIn
Healthcare sustainability challenge. Demographics, economic crisis, provider architecture, and cloud architectural patterns as inspiration.
Conference · AFCRO 2026 · Qualees · YouTube
TweenMe pipeline applied to the OCTOPUS study. Population synthesis, TSTR validation 95.2%, counterfactual simulator. ISPOR 2026 top 5%.
Conference · ISPOR Europe 2025
Validated TVAE pipeline for high-fidelity synthetic NSCLC cohort generation. ML Utility 0.907, correlation 0.935. OCTOPUS study, n=184.
Conference · ISPOR Europe 2025
TweenMe Single/Best/Ensemble: Bayesian optimisation OPTUNA, SMOTE, stacking. Accuracy 99.50%, ROC AUC 99.97% on diabetes.
Conference · AFCRO 2025 · YouTube
AI in clinical trials: methodological approaches, synthetic control arms, deployment in a demanding regulatory framework.
Conference · PHUSE 2024 · Paper CM05
GANs, VAEs, Transformers applied to drug discovery, clinical data automation, precision medicine and decentralized trials. 16 pages.
Conference · AFCRO 2024 · YouTube
TweenMe presentation at AFCRO 2024. Data pipeline architecture, dimensional enrichment, first clinical validation results.
Conference · AbbVie · YouTube
AbbVie symposium for oncology clinicians. AI demystified, clinical research use cases, and computational haemato-oncology.
Video · VivaTech 2019 · Microsoft H&LS · YouTube
VivaTech 2019 on AI trajectory in medicine and requirements for industrial deployment. First TweenMe/PREDICARE formulation.
Article — Position paper · Twingital Institute
Of four digital sovereignty layers, only the energy layer is allocated, not composed. AI enters macro-industrial governance.
Article — Position paper · Twingital Institute
Seven layers, three verdicts: arbitration doctrine of composite sovereignty for AI in regulated European environments. Silicon changes political nature.
Article — Position paper · Twingital Institute
AI kilowatt-hour allocation is triply distributed (market, contracts, local blockages) without deliberative procedure. Conditions for a sectoral protocol.
Exploration · Twingital Institute
The CVE-2026-31431 (Copy Fail) case on WSL2 kernel 6.6.87.2 as empirical terrain for the promotion gate between technical artifact and normative dependency.
Exploration · Twingital Institute
JEPA and latent predictive architectures: geometry of predictability, anti-collapse constraints, the unmet threshold of biographical memory.
Exploration · Twingital Institute
ToxTwin V2.4: endpoint tri-router, enriched Ames ISS and hERG ChEMBL corpora. Tox21 mean AUC 0.898, Ames 0.853, hERG 0.800. Lessons from a controlled cycle.
Exploration
Full toxicological report on doxorubicin via ToxTwin V1.3: 14 Tox21 endpoints, Ames, hERG, mechanistic interpretation, and literature concordance.
Exploration · Twingital Institute
ToxTwin V2.3 technical synthesis: 14 endpoints, AUC 0.867, corrected biases, IR calibration, composite AD, V2.4 plan.
Exploration · Twingital Institute
ToxTwin V2.3 user guide for toxicologists and researchers. SMILES, 14 endpoints, calibrated scores, applicability domain, FAQ.
Reference framework · Twingital Institute
Why agentic AI is fundamentally an enterprise architecture problem. Five structural functions of EDA for situated, auditable, governable AI agents.
Article — Position paper · Twingital Institute
Gap between massive deployment and real adoption of Microsoft 365 Copilot. 3.3% penetration, negative NPS, generative shadow IT.
Exploration · Twingital Institute
ToxTwin FastAPI API documentation. Aspirin and ibuprofen validation tests, differential NR-PPAR-gamma signal, ScoreResponse format.
Exploration · Twingital Institute
Dreyfus, Searle, and the language/representation confusion. Explicit, implicit, and pipeline encoding. The data scientist layer.
Exploration · Twingital Institute
Biological transduction vs mediated encoding. Symbol grounding problem. Molyneux's problem, Mary's argument. Three grounding levels.
Exploration · Twingital Institute
Memory as polycentric graph. Biographical edge vs statistical edge. World models: decisive advance but insufficient threshold.
Exploration · Twingital Institute
Full article (28 pp.). Encoding, transduction, stratified grounding, polycentric memory graph, biographical edge, world models.
Exploration · Twingital Institute
Full ToxTwin workflow on an unknown candidate out of domain (Tanimoto 0.153). SR-p53 (32.6%) and SR-ARE (25.5%) signals.
Exploration · Twingital Institute
End-to-end ToxGNN-V1 pipeline (Mean AUC 0.857 on Tox21). Medallion architecture, NER extraction via Phi-4 14B, MDR and EU AI Act.
Article — Position paper · Twingital Institute
Why most Bayesian implementations in health AI are frequentism dressed up as a probabilistic graph. Calibration, drift, and the HR argument.
Article — Position paper · Twingital Institute
AI explainability in medicine is a relational predicate, not an absolute. Measured, audited, monitored reliability over illusory transparency.
Article — Position paper · Twingital Institute
Analysis of AI's physical constraints: energy, water, data, silicon. $200B in GPUs, <10% adoption, Jevons rebound effect.
Article — Position paper · Twingital Institute
Six-dimension eligibility framework: decision topology, time balance, uncertainty reduction, coupling, TCO, reversibility.
Article — Position paper · Twingital Institute
Structural analysis of AI-induced clinical cognitive load: techno-solutionism, collective decision-making, strained IT departments, TweenMe encapsulation.
Exploration · Twingital Institute
GVI/PCI analysis (MIT CPO) of the French sovereign cloud market: hyperscalers, SecNumCloud, territorialised hybridisation, risk matrix.
Article — Position paper · Doctrinal position · LinkedIn
Impressive technical progress does not exempt from economic and clinical proof. The missing independent health-economic evidence for AI in healthcare.
Exploration · Le Coin du Polymathe · LinkedIn
Answers to fundamental quantum computing questions: storage, computing basis, memory architecture and RSA decryption.
Exploration · Le Coin du Polymathe · LinkedIn
The quantum stability challenge: bosonic cat-qubits, Amazon's Ocelot chip and Alice & Bob's TomCat architecture.
Exploration · Le Coin du Polymathe · LinkedIn
Majorana fermions — particles that are their own antiparticle — as candidates for fault-tolerant topological qubits.
Exploration · Le Coin du Polymathe · LinkedIn
Comparison of qubit architectures: superconducting, trapped ions, topological, neutral atoms, photonic, silicon spin.
Exploration · Le Coin du Polymathe · LinkedIn
Quantum mechanics fundamentals: qubits, superposition, entanglement, Bloch sphere, observables and wave function collapse.
Article — Position paper · Twingital Institute · January 2025
AI as research accelerator in bio-industries. Data quality, HDLSS, cybersecurity, AlphaFold v3 limitations.
Article — Position paper · Twingital Institute
Why the patient digital twin concept lacks defensible scientific content — and why the answer is a generator.
Article — Position paper · Twingital Institute
From buzzword to operational discipline. Analytical framework, maturity models, ontological intelligence, cybersecurity and TweenMe.
Article — Position paper · Twingital Institute
Critical analysis of agentic AI frameworks. Historical roots, structural novelties, persistent challenges, and the cybersecurity blind spot.
Exploration · LinkedIn · January 2024
What emergency medicine and philosophy of mind say about consciousness — and its implications for AI.
Exploration · Twingital Institute
Perceptrons, transformers, emergent complexity through massification. The brain is fractal, not a lasagna dish.
Exploration · LinkedIn · April 2020
COVID-19 excess mortality in nursing homes: geriatric frailty, polypharmacy, ICU limits, palliative care ethics.
Exploration · LinkedIn · March 2020
Physiopathology of SARS-CoV-2: ACE2 binding, TMPRSS2 activation, cytokine storm, ARDS, sepsis, and therapeutic approaches (March 2020).
Exploration · LinkedIn · March 2020
RT-PCR mechanics, ELISA immunological tests, sensitivity/specificity trade-offs, and why massive testing alone cannot contain a pandemic.
Exploration · LinkedIn · March 2020
Fundamental virology refresher: viral structure, envelope, genetic payload, replication cycle, mutations and recombination of SARS-CoV-2.
Exploration · Twingital Ventures
Analysis of frugal innovation in a VUCA world: short cycles, trial-and-error culture, MVP, and the banking industry case.
Conference · AFCRO 2026 · Qualees · YouTube
TweenMe pipeline applied to the OCTOPUS study. Population synthesis, TSTR validation 95.2%, counterfactual simulator. ISPOR 2026 top 5%.
Conference · AFCRO 2025 · YouTube
AI in clinical trials: methodological approaches, synthetic control arms, deployment in a demanding regulatory framework.
Conference · AFCRO 2024 · YouTube
TweenMe presentation at AFCRO 2024. Data pipeline architecture, dimensional enrichment, first clinical validation results.
Conference · AbbVie · YouTube
AbbVie symposium for oncology clinicians. AI demystified, clinical research use cases, and computational haemato-oncology.
Video · VivaTech 2019 · Microsoft H&LS · YouTube
VivaTech 2019 on AI trajectory in medicine and requirements for industrial deployment. First TweenMe/PREDICARE formulation.
Is it economically sustainable? Is it governable over time? If any of these questions remains open, that's where the work begins.