01 / INTELLIGENCE
Curova IQ Engine is the governed intelligence layer that learns from consented, structured, longitudinal events across the healthcare journey — and ships its outputs as reviewable, attestable, human-in-the-loop artefacts.
02/DECISION TRAIL GRAPH
The Decision Trail Graph turns the journey into a reviewable object. Each node carries provenance, reasoning, and clinical sign-off — and every edge between nodes is queryable.
03/GOVERNED CARE LEDGER
The Governed Care Ledger is the durable, cryptographically chained history beneath the Decision Trail Graph. Patients can replay it at the scope of their consent. Clinicians can audit it across encounters. Regulators can attest to its integrity without reading PHI.
The ledger is the source of truth Curova IQ reasons over.
04/HUMAN-IN-THE-LOOP
HITL in Curova is not a checkbox at the end of a flow. It’s wired into every AI call site as a typed primitive.
Every recommendation ships with a confidence band. Low-confidence outputs route to escalation before they reach the patient.
Sign-off is a typed object with named clinician, timestamp, modifications, and the underlying AI artefact attached.
Clinicians can override any AI output. The override is captured as a first-class event with the rationale, feeding evals.
Encoded triggers — red flags, contraindications, low-confidence — route to a named on-call clinician with measured response.
05/MULTILINGUAL REASONING
Multilingual is not a thin translation layer. It is a first-class reasoning surface, evaluated per language and per task.
Health-tuned reasoning across Hindi, Tamil, Telugu, Bengali, Marathi, Kannada, Malayalam, Gujarati, Punjabi, Odia, Assamese, and more.
Patients speak. The substrate transcribes, normalizes, and routes — preserving original utterance for audit and review.
Clinical-eval rails track precision, recall, and safety per language and per task. No average hides a failure mode.
06/EVAL & RED-TEAM
Curova’s evaluation rails are part of the substrate, not a marketing surface. Every model version ships with attached evaluation artefacts.
Models are evaluated against versioned clinical benchmarks on every release — and silently in production against ground truth.
Red-team suites probe for unsafe completions, jailbreaks, and edge-case clinical reasoning. Failures route into the dev loop.
Decision-trail nodes carry model identity. Outcome data attributes back to the version that informed the decision.
07/THE PILLARS
Every interaction — symptom, note, lab, prescription, claim — captured as a node with provenance, reasoning, and clinical sign-off.
An append-only ledger of care events, cryptographically chained, queryable by patient, clinician, and regulator.
Structured sign-off, confidence-banded outputs, and clinician override surfaces wired into every AI call site.
Health-tuned models that reason in 22+ Indian languages — and the substrate to evaluate them per language, per task.
Continuous evaluation against clinical benchmarks and adversarial probes; failures route into the development loop.
Care does not move in one system. Curova IQ subscribes to events from every module, structured at the source.
PARTNER
We work with research partners, clinical leaders, and integration partners ready to put intelligence on a governed substrate.