Feature · DeepSensi PBC, Dover, Delaware · July 2026 · [email protected]
Modern medicine has a memory problem. A patient is examined, diagnosed, treated — and then, administratively speaking, forgotten. If science later discovers that her medication interacts dangerously with another, or that her lab pattern from three years ago matches a newly described syndrome, no mechanism exists to tell her. The knowledge arrives; the connection never does. DeepSensi, the verified clinical-intelligence system built by Tomasz Jan Gomoła, was designed around a blunt refusal of that amnesia — and its patient-facing machinery, detailed in the company's newly published technical papers, shows what medicine looks like when it is architected to remember.
Every patient in the system has a longitudinal digital twin — a living model of their health trajectory rather than a static file. The twin is what makes a capability called retroactive alerting possible: when new medical knowledge enters the system's verified evidence base — a freshly published interaction, a revised guideline, a newly described risk pattern — DeepSensi automatically re-examines the patients whose histories it already knows. If the new knowledge changes the picture for a patient seen last month or last year, the treating physician is alerted. The physician no longer needs to remember every patient; the system never forgets one.
The twin also watches forward. Continuous monitoring of biomarker trends can surface a quiet drift — the sub-clinical inflammation, the slowly climbing pressure — before it becomes an emergency-room visit, triggering proactive, personalized outreach in the patient's own language. And that language matters more than it sounds: every patient-facing report is generated as a plain-language health summary and validated with readability metrics before it is sent, so that a person in crisis receives an explanation, not a wall of jargon.
DeepSensi is already known for LIMBO, its structured "I don't know" — the protocol that made an honest declaration of uncertainty into a first-class clinical output. The patient-facing complement is a capability the papers call active sensing: when the diagnostic picture is ambiguous, the system does not merely admit uncertainty — it computes which single examination or laboratory test would most decisively resolve it, and puts that recommendation in front of the physician. For the patient, the difference is concrete: fewer shotgun test batteries, less time in diagnostic limbo, and a clear answer to the question every anxious person asks — what do we do next?
The most unexpected capability hides inside the system's most mundane feature. DeepSensi's autonomous scribe writes structured clinical notes so physicians don't have to. Multiplied across clinics, those anonymized notes become something else entirely: a real-time syndromic map. The Global Emergency Response Network continuously scans de-identified documentation across participating sites, and when a cluster of similar presentations begins forming in one geographic region — the early signature of an outbreak — it flags physicians and can alert health authorities. No new bureaucracy, no separate reporting forms: the early-warning system is a by-product of doctors simply doing their documentation, with privacy preserved by the same anonymization architecture that governs everything else in the platform. Public-health surveillance has historically depended on hospitals remembering to report; here, the radar is always on.
For patients who have exhausted standard-of-care, the system deploys its most human module: Golden Horizon, a compassionate-access engine that scans the world's clinical-trial registries and then sends an autonomous agent to advocate — contacting trial providers, following up on a fixed schedule, re-scanning when the landscape changes, and reporting back to the patient in plain, warm language that explicitly tells them: you don't need to do anything; we are handling this. It is free for patients by architectural mandate — a zero-cost constant hardcoded into the system so that no future business decision can revoke it.
Each of these capabilities exists elsewhere in fragments — patient portals, screening reminders, pharmacovigilance databases, trial-matching websites. What has not existed is their integration under a single verification architecture, in which every alert, every summary, and every match must pass the same anti-hallucination cascade that bounds the system's error rate, and every action is sealed into a court-grade audit trail. A reminder is only as good as the evidence behind it; an epidemic alert is only as good as its false-positive discipline. DeepSensi's wager is that patient-facing features earn trust only when they inherit the safety mathematics of the whole system — and that a health system which remembers its patients, watches over them between visits, and advocates for the ones running out of time is not a luxury feature of medicine's future. It is what medicine was always supposed to do, restored by an architecture that finally has the memory for it.
Technical documentation: www.deepsensi.com/papers · Press and auditor access: [email protected] · DeepSensi PBC is a Public Benefit Corporation; free trial matching, open research, and the open DSS safety standard are written into its charter.