From reasoning to continuous care
Wave 1 built 38 products that reason clinically. Wave 2 is a different kind of system — ten applications built around one shared patient state, closed clinical loops, and an evidence ledger that proves care changed the outcome.
Can we reason clinically at population scale?
A Wave-1 product is done when it reasons correctly.
Can we hold one patient's state continuously, act on it across every setting, and prove it changed the outcome?
A Wave-2 product is done when it writes to shared state, closes a loop, and emits evidence.
The distinction changes what “done” means. Done is no longer a correct answer — it is a write to shared state, a closed loop, and an outcome-trackable event.
The cascade
Every Wave-2 application reads from and writes to one continuous patient state, and everything that acts emits evidence downstream. The cascade is the product.
AKHANDA and VYUHA are working names; PRAMANA is under review. Dashed boxes are reasoning and specialty surfaces that read and write the shared state.
The ten applications
Ten applications, 1,024 new knowledge bases, four build phases. Three names are provisional pending a final naming decision.
Patient State / Clinical Graph — the one place the patient exists. Sanskrit akhaṇḍa, “unbroken, indivisible”.
Dependencies: none — foundational; blocks everything downstream.
Read the blueprint →Outcomes & Evidence Ledger — proves care changed the outcome. Every Wave-2 product emits to it.
Dependencies: AKHANDA events.
Longitudinal care orchestration — chronic disease trajectories and goals across settings.
Dependencies: AKHANDA.
Medication intelligence — recommends and blocks, never prescribes. Extends the built GUARDIAN.
Dependencies: AKHANDA; extends the built GUARDIAN.
Emergency department intelligence — the ED loop of the cascade.
Dependencies: AKHANDA; integrates SHAYYAA, PRAANA, SUSHRUT.
Nursing workforce intelligence — the nursing layer of the cascade.
Dependencies: AKHANDA, SHAYYAA.
Home & virtual care — care delivery beyond the hospital walls.
Dependencies: composes SETU, Aegis, ARES, Twin, SHRAVANI + JEEVANA.
Hospital operational twin — command centre. Sanskrit vyūha, “strategic formation”.
Dependencies: AAROGYA, SHAYYAA, PRAANA, SUSHRUT, ABHAYA.
Federated learning across sites — models travel, data does not.
Dependencies: PRAMANA.
Patient companion — the patient-facing surface of the cascade.
Dependencies: AKHANDA, JEEVANA, GUARDIAN+.
AKHANDA and VYUHA carry working names pending a final naming decision. PRAMANA (Sanskrit pramāṇa, “proof, means of valid knowledge”) is under review, with SAKSHYA (sākṣya, “evidence, testimony”) as the alternative.
How Wave 2 is built
Knowledge bases come first
Every Wave-2 product is KB-first: knowledge bases are authored, clinically reviewed, and frozen before the engine that consumes them. Every KB carries a claim grade that governs downstream hedging language — frontier-hypothesis content may never appear in a clinician-facing recommendation without an explicit uncertainty banner.
The layer taxonomy — uniform across all ten products
| Layer | Name | Contains |
|---|---|---|
| L0 | Ontology | Entities, codes, value sets, units, identity rules |
| L1 | Domain Knowledge | Clinical facts, guidelines, physiology, thresholds |
| L2 | Reasoning Templates | Decision trees, scoring, trajectory models |
| L3 | Workflow | State machines, role handoffs, escalation ladders |
| L4 | Integration | Contracts with other QuantVeda products, FHIR mappings |
| L5 | Safety & Governance | Prohibitions, red lines, audit rules, consent gates |
Scores are computed by code. Narratives are generated by the model. These never swap.
A language model never produces a risk score, a dose, a triage category, or a probability. It receives pre-computed values and writes the clinical prose around them. This is mandatory for every Wave-2 product; a violation is a P0 defect.
Event-driven, evidence-emitting
All inter-product communication is event-driven, never direct invocation. Every Wave-2 product must emit to PRAMANA: a product that takes a clinical action and emits no outcome-trackable event is incomplete.
Seven universal safety rails
Every rail is implemented as code, not prompt instructions, and applies to all ten applications.
Human-in-the-loop gateway
No Wave-2 product issues a clinical instruction to a patient without a licensed clinician's approval event recorded in AKHANDA.
No autonomous prescribing
GUARDIAN+ recommends and blocks; it never prescribes.
Escalation is uninterruptible
Any deterioration signal must reach a human. No queue, rate-limit, or quiet-hours rule may suppress a red-tier alert. SAATHI's quiet hours (22:00–07:00) apply to nudges only, never to alerts.
Consent is per-purpose and revocable
Care delivery, quality improvement, research, and commercial RWE are four separate consents. SANGAMA may only train on the third and fourth.
Provenance on every assertion
Every clinical claim surfaced to a user resolves to a KB id and version. Unattributable output is suppressed.
Degraded mode is defined, not emergent
Every product declares what it does when its dependencies are unavailable. The default is fail-safe: reduce function, never guess.
Paediatric, pregnancy, and geriatric gates
Any dosing or physiological threshold checks life-stage before emitting.
Build sequence: Phase A → D
Foundational
- 1 · AKHANDABlocks everything downstream
- 2 · PRAMANADepends on AKHANDA events
Clinical loops
- 3 · JEEVANADepends on AKHANDA
- 4 · GUARDIAN+Extends built GUARDIAN; depends on AKHANDA
- 5 · ABHAYADepends on AKHANDA; integrates SHAYYAA / PRAANA / SUSHRUT
Workforce and setting
- 6 · SEVADepends on AKHANDA, SHAYYAA
- 7 · GRIHA VYDYAComposes SETU / Aegis / ARES / Twin / SHRAVANI + JEEVANA
- 8 · VYUHADepends on AAROGYA, SHAYYAA, PRAANA, SUSHRUT, ABHAYA
Scale and surface
- 9 · SANGAMADepends on PRAMANA
- 10 · SAATHIDepends on AKHANDA, JEEVANA, GUARDIAN+
What “done” means for every product
No Wave-2 product ships until every line below holds.