See It In Action
What You'll See
Meet the AI: Deutsch & Popper
Our agents are named after two philosophers whose epistemology shapes our approach. Deutsch (TA1) honors David Deutsch and the "hard-to-vary" principle. Popper (TA2) is named for Karl Popper, father of falsificationism.
Together: AI that doesn't just predict — it generates explanations that can be tested and refuted.
Population Dashboard — Massive Scale
Deutsch and Popper manage cardiovascular care for over 4,000 patients with a team of just 3 clinicians.
Tonight, they'll help Maria, who lives 90 miles from the nearest cardiologist.
Maria's 2 AM Emergency
Maria wakes with shortness of breath and swollen ankles. She opens the Regain app — her wearable data syncs automatically, showing a 5-pound weight gain and elevated heart rate. She describes her symptoms.
Her health record loads via FHIR: labs, medications, problem list, vitals — all aggregated and ready for analysis.
Hard-to-Vary Differential Diagnosis
Deutsch generates hypotheses, then adversarially attacks each one. The Verifier critiques using Hard-to-Vary scoring — measuring how tightly each explanation fits the evidence.
- • When uncertain, Deutsch explicitly marks hypotheses as "insufficient data" rather than rejecting them
- • Weak hypotheses are eliminated; strong survivors become tentative diagnoses
- • The Universal Domain Composer activates CVD, Nutrition, Exercise, and Mental Health subsystems simultaneously
When uncertain, the system says "I don't know." Full transparency about what it knows — and doesn't know.
Popper's Safety Validation
Popper validates every recommendation: protocol reference present, dosing within parameters, no contraindications detected.
- • Approve — medication within protocol, decision: approved
- • Hard-stop — dose exceeds maximum parameters
- • Request info — data is stale, need more information
- • Route to clinician — new medication class proposed, requires human approval
The system knows when to act autonomously — and when human judgment is required.
Full Transparency & FDA Ready
Every decision is fully traceable. Evidence references, HTV score breakdowns, key unknowns disclosed. Human clinicians see exactly why Deutsch decided on a diagnosis or made a recommendation.
For FDA review, complete performance reports are exported — automatically de-identified and formatted for regulatory submission.
We're not proposing to build this. We're showing a working prototype.