What if there was a solution…
Cognitive Medical OS
Difficult Cases Engine.
Enter DeepSensiYou are viewing DeepSensi United States. It looks like you are in .
SYSTEMS BIOLOGY COGNITIVE ENGINE // DIFFICULT CASES ENGINE
Point tools automate tasks. DeepSensi™ replaces the operating conditions of medicine itself: how a diagnosis is reasoned, verified, recorded, remembered, paid for, and defended. Opening in stages.
DETERMINISTIC CLINICAL SAFETY // IEC 61025 BOUNDED UNCERTAINTY
Conventional AI hallucinates when it runs out of data; DeepSensi activates LIMBO. DeepSensi does not guess—it deduces. When the clinical record falls short of definitive proof, the engine halts speculation, declares structured uncertainty, and pinpoints the single decisive investigation that transforms ambiguity into proven evidence.
CLINICIAN COGNITIVE EXOSUIT // SPECIALIST CONSILIUM ON DEMAND
One physician now carries the provable certainty of an entire hospital: an ambient scribe, an exosuit that shares the load, and a verified safety architecture under every conclusion. A decentralized teaching hospital, on demand.
PEER CLINICIAN NETWORK // WP-003
By joining the Hyper Consilium, you become a verified node in the world’s most elite diagnostic network.
HEALTH SYSTEMS DEPLOYMENT // INPATIENT EHR & SAFETY GOVERNANCE
Department-grade intelligence in the systems you already run: EHR plug-and-play, AI risk you can insure against a quantified worst-case bound, and EU AI Act readiness by architecture.
DECENTRALIZED CLINICAL RESEARCH // PopSIM™ SYNTHETIC COHORTS
Every failed trial died the same death: the right patient, found too late. DeepSensi matches at the exact moment of diagnosis, simulates the cohort before a site opens, and accelerates decentralized clinical trials with synthetic control arms.
GLOBAL SURVEILLANCE // GERN PATHOGEN & EPIDEMIC RADAR
GERN turns ordinary clinical notes into DeepSensi’s exclusive epidemic radar: physician-level early warning that crosses borders. Open, royalty-free, auditable, private by architecture.
AUDITABLE BY DESIGN // IEC 61025 FAULT TREE ANALYSIS
A quantified error bound, derived by fault tree analysis, liability partitioned per decision, and 100 percent audit coverage of every assertion. And the missing piece: an open, royalty-free certification standard that maps onto the EU AI Act. The measuring stick you were asked to invent already exists.
SOVEREIGN INFRASTRUCTURE // AIR-GAPPED ON-PREMISE HOSPITAL NODES
From a clinic gateway to a fully air-gapped hospital node, DeepSensi runs where the patient is. No cloud required. Nations keep their data; field teams keep their radar.
PROSPECTIVE PROTOCOL // SILENT-DS™ MULTI-SITE TRIAL
SILENT-DS: a multi-site, silent-mode prospective study. Pre-registered before the first patient, independently adjudicated, reported whatever the results. Hospitals join as named study sites.
What DeepSensi is. DeepSensi is the World’s Preeminent Difficult Cases Engine, engineered specifically for intractable diagnostic odysseys, complex multi-system illness, and the inter-specialty white space where conventional medicine stalls. Governed by the open DSS-001 Platinum Standard (Gomola Framework), it executes deterministic cross-domain physiological deduction with a formally analysed safety architecture across 60+ biological databases. Physician-led sovereign intelligence and mathematically verified evidence.
Safety architecture designed to a SIL-4-class target, quantified by fault tree analysis under IEC 61025 across 23 independent barriers. Not certified by an accredited third party. How we measure it: deepsensi.com/safety-architecture
HOW DEEPSENSI WORKS // ONE CASE, FOUR STEPS
Every case enters the same way, under every plan: through My DeepSensi on a desktop or a phone, or through an Edge Node in the room. Identity is removed on the device, by the device's own AI, before anything leaves it.
See My DeepSensi
The Difficult Cases Engine and the Systems Biology Cognitive Engine work the case across every specialty and the white space between them, on a formally analysed safety architecture. Around them stands a suite of functions for the physician, the hospital and pharma that until now were scattered across separate tools or did not exist at all, among them the Clinician Exosuit and a new economics of medicine built for the age of AI.
Explore the platform
A family's Precision Suite, a physician's practice, a hospital ward, a registry: each plan delivers the result its own way. What is inside never changes: every hypothesis with its sources, every step auditable.
Plans and deployments
An ambient scribe and an empathic assistant that remembers the person and talks like one, as if the physician never left the bedside. Thin, thick or air-gapped.
Meet the Edge NodeAnd it learns. Cohort learning carries what one resolved case teaches to the next, across the whole network, while the records themselves never move. Patent-pending.
CLINICAL BENCHMARK // 301 NEJM CPC EVALUATED CASES
On the publicly available NEJM Clinicopathological Conference cases, the diagnoses that stump specialists, DeepSensi does not just score well. It refuses to miss the ones that matter. In clinical systems, state is not declared by code. State is a function of proven evidence.
Diagnostic accuracy on NEJM Clinicopathological Conference cases
Test conditions. NEJM Clinicopathological Conference (CPC) cases, text only (no vision), in a double-blinded test. DeepSensi was not trained on these cases: it reaches each diagnosis by medical deduction. The published results below it come from another protocol and are shown with their source.
DeepSensi test: double-blinded, text only (no vision). Comparison models run zero-shot on the same 100 cases.
Published results on NEJM CPC cases: Microsoft SDBench, sequential diagnosis protocol.
Missed critical: the correct diagnosis is not among the top three. Sources, per model: DeepSensi Cognitive Engine: DeepSensi development benchmark, double-blinded, 301 NEJM CPC cases (2014 to 2023), text only, DeepSensi “FAST Challenger” configuration. Gemini 3.8 Flash (high reasoning): DeepSensi zero-shot comparison study, double-blinded, 100 NEJM CPC cases, single model, model gemini-3.8-flash, high reasoning, 18 September 2026. Claude Opus 4.8: DeepSensi zero-shot comparison study, double-blinded, 100 NEJM CPC cases, single model, model claude-opus-4-8, September 2026. Gemini 2.5 Pro: DeepSensi zero-shot comparison study, double-blinded, 100 NEJM CPC cases, single model, model gemini-2.5-pro, September 2026. Claude Fable 5: DeepSensi zero-shot comparison study, double-blinded, 100 NEJM CPC cases, single model, model claude-fable-5, September 2026. MAI-DxO, o3, Gemini 2.5 Pro, Claude 4 Opus, GPT-4o and physicians: Nori et al. 2025, “Sequential Diagnosis with Language Models”, Microsoft AI, arXiv:2506.22405: 304 NEJM CPC cases (2017 to 2025), sequential protocol in which the model asks questions and orders tests before giving one final diagnosis; MAI-DxO in its best configuration; Gemini 2.5 Pro and Claude 4 Opus as off-the-shelf baselines (Figure 8); physicians on the 56-case test set.
| System | Protocol | Cases | Top-1 | Top-3 | Missed critical | Source |
|---|---|---|---|---|---|---|
| DeepSensi Cognitive Engine | FAST Challenger | 301 | 93.3-96.7% | 100% | Zero | DeepSensi development benchmark, double-blinded |
| Gemini 3.8 Flash (high reasoning) | Zero-shot | 100 | 82.0% | 91.0% | 9.0% | DeepSensi zero-shot comparison study, double-blinded |
| Claude Opus 4.8 | Zero-shot | 100 | 70.0% | 83.0% | 17.0% | DeepSensi zero-shot comparison study, double-blinded |
| Gemini 2.5 Pro | Zero-shot | 100 | 58.0% | 82.0% | 18.0% | DeepSensi zero-shot comparison study, double-blinded |
| Claude Fable 5 | Zero-shot | 100 | 39.0% | 55.0% | 45.0% | DeepSensi zero-shot comparison study, double-blinded |
| Microsoft MAI-DxO (orchestrating o3) | Sequential | 304 | 85.5% | Not measured | Unavailable | Nori et al. 2025, arXiv:2506.22405 (SDBench) |
| OpenAI o3 | Sequential | 304 | 78.6% | Not measured | Unavailable | Nori et al. 2025, arXiv:2506.22405 (SDBench) |
| Gemini 2.5 Pro | Sequential | 304 | 68.8% | Not measured | Unavailable | Nori et al. 2025, arXiv:2506.22405 (SDBench) |
| Claude 4 Opus | Sequential | 304 | 68.4% | Not measured | Unavailable | Nori et al. 2025, arXiv:2506.22405 (SDBench) |
| OpenAI GPT-4o | Sequential | 304 | 49.3% | Not measured | Unavailable | Nori et al. 2025, arXiv:2506.22405 (SDBench) |
| Practising physicians, unaided | Sequential | 56 | 19.9% | Not measured | Unavailable | Nori et al. 2025, arXiv:2506.22405 (SDBench) |
Performance results. The figures in this section were measured on the diagnostic device configuration, DeepSensi Cognitive Medical OS, in its development phase: an internal measurement that is not independent of the Company, on 301 NEJM Clinicopathological Conference cases (2014 to 2023), with single frontier models run zero-shot as comparators (cases per row as shown); published results of other systems are cited with their source. Regulatory status: deepsensi.com/regulatory
HEALTHCARE EPIDEMIOLOGY // ECONOMIC COST OF DIAGNOSTIC ERROR
Every year the system pays, in full, for the exact problems this platform removes. The money is already being spent. It is simply being spent on the damage.
Sources: Newman-Toker et al., BMJ Quality & Safety 2023; Society to Improve Diagnosis in Medicine; AMA / Shanafelt et al., Annals of Internal Medicine; industry trial-recruitment and delay-cost studies; Cutler & Summers, JAMA 2020; Nori et al. 2025 (SDBench, sequential protocol).
EVIDENTIARY ARCHITECTURE // VERIFIABLE CLINICAL PROOF RECORD
The worst-case probability of an undetected hallucination reaching a conclusion: one in 309,600 assertions, derived by fault-tree analysis per IEC 61025, not asserted by marketing. Under nominal conditions: one in 5.9 million.1
Every conclusion ships with a court-grade evidentiary record: each step attributed to AI, physician, or rule, cryptographically sealed, tamper-proof even against the operator.
A specialist counterpart for every discipline of medicine deliberates adversarially under independent judges, and physicians join as scored nodes. The reach of a teaching hospital, upon evaluation.
The diagnosis that recruits. The instant a phenotype is recognized, the patient is matched to a clinical trial or a compassionate-access route. Enrollment starts at diagnosis, not months later.
SYSTEMS BIOLOGY // THE INTER-SPECIALTY WHITE SPACE ENGINE
Modern medicine fragments human biology into isolated specialties: neurology monitors the seizures, immunology tracks cytokines, endocrinology tests hormones, genetics isolates VUS variants. When standard protocols hit a wall, the failure is not lack of data. It is the blind spot between disciplines: the Inter-Specialty White Space.
Over 80% of drug-resistant, undiagnosed, and multisystem conditions inhabit the unmapped gaps between clinical departments. DeepSensi treats cross-disciplinary interactions not as confusing edge-cases, but as the primary mathematical manifold where root etiology hides.
Real-time cross-referencing of urinary organic acids (OAT 75+), microbial toxins (HPHPA, 4-cresol), WGS Trio genomics, receptor autoantibodies, and pharmacokinetic tensors. One unified computational consilium instead of ten siloed doctor appointments.
Benchmarked across 301 NEJM CPC cases. When single-specialty medicine exhausts standard guidelines, the White Space Engine finds the causal mechanism.
CLINICAL STAKEHOLDER MATRICES // PRACTICE IMPACT
CLINICAL PRACTICE // ATTENDING PHYSICIANS & SPECIALISTS
A full specialist consilium in seconds, and a court-grade record behind every decision you make. DeepSensi is your clinical exosuit: one physician now wields the provable certainty of a full specialist consilium.
What changes for you
PRECISION MEDICINE // PATIENTS & PEDIATRIC FAMILIES
A university-grade second opinion for your case, in your language, in seconds, with every conclusion backed by evidence and never a guess. And medicine that remembers you: an alert can reach you years after a visit, when a new finding finally connects the dots.
What it means for you
HEALTH SYSTEMS // UNIVERSITY HOSPITALS & CLINICAL NETWORKS
Department-grade intelligence with zero new headcount. EHR plug-and-play, insurable AI risk, EU AI Act ready by architecture, and documentation that writes and codes itself.
Deploy in your department
EPIDEMIOLOGICAL SURVEILLANCE // GERN PATHOGEN RADAR
GERN turns ordinary clinical encounters into DeepSensi’s exclusive epidemic radar: physician-level early warning that crosses borders, built from the notes doctors already write. Open, royalty-free, auditable, private by architecture.
The exclusive epidemic radar
BIOPHARMA ENTERPRISE // PopSIM™ IN SILICO COHORTS
Verified-evidence discovery, population-scale simulation, and real-time trial matching at the exact moment of diagnosis, on four continents.
Attack the denominator
REGULATORY SPECIFICATION // IEC 61025 ERROR BOUND
The first clinical AI with a quantified error bound, derived by fault tree analysis, liability partitioned per decision, and 100 percent audit coverage of every assertion. Governed by the Gomola Framework, an open, royalty-free certification standard.
The DeepSensi Standard
PEER-REVIEWED FOUNDATION // WHITE PAPERS & PROSPECTIVE EVIDENCE
We publish the proof, the standard, and the audit. We do not publish the blueprint.
PROSPECTIVE PROTOCOL // STARD-AI & SPIRIT-AI ADJUDICATED
A prospective, multi-center, observer-blinded study of diagnostic accuracy and safety across complex inpatient admissions, run in silent mode. Designed under STARD-AI / SPIRIT-AI to test DeepSensi’s diagnostic concordance for non-inferiority and then superiority, including missed critical diagnoses and inter-specialty root causes. Hospitals participate as named co-investigator study sites.
DIAGNOSTIC UTILIZATION // SEQUENTIAL PROTOCOL HEALTH ECONOMICS
In the same NEJM CPC case family, this is what reaching one diagnosis costs today, in ordered diagnostic tests (sequential protocol, Nori et al. 2025):
Sources: Microsoft MAI-DxO 85.5%, OpenAI o3 78.6%, OpenAI GPT-4o 49.3% and practising physicians, unaided, 19.9%: Nori et al. 2025, arXiv:2506.22405 (SDBench, sequential protocol).
Different protocol, honestly stated: SDBench orders tests sequentially; DeepSensi reasons from the presented record, and when evidence is genuinely missing, LIMBO names the single test that settles the case, not a battery. That is the inversion: no test cascade to pay for, and a consilium engineered to run on commodity infrastructure at a per-case cost hundreds of times below the bills above, under one percent of them, stated deliberately as a bound, not as “free.” Certainty at a fraction of what the market already pays for guesswork; deployment pricing is quoted per pilot.
DEEPSENSI // COUNTRY BY COUNTRY
RESOLVE™ pre-enrollment is open in most countries. Enrollment opens in stages, the United States first, each under independent ethics (IRB) review; countries follow in order of demand.
| Country or region | RESOLVE registry |
|---|---|
| USUnited States | RESOLVE OpeningIRB submission in preparation, the first country to open |
| CACanada | Next |
| AUAustralia | Next |
| +European Economic Area | Not yet open |
| +United Kingdom, Switzerland, mainland China, Thailand, Türkiye | Not available |
| +Every other country | Pre-enrollment opencountries open in order of demand |
