KINETICA AI

Clinical AI that hallucinates is dangerous. Clinicians need AI that knows its limits and stops when uncertain.

ReAct agent with human-on-loop architecture. 9-node LangGraph graph where the clinician decides at every gap. ALMA ethical framework evaluates every output before it reaches the patient.

1,880 RAG chunks, deterministic safety layer (regex + cosine, zero LLM calls in evaluation), zero autonomous escalation. EU AI Act compliant by design.

Patients with complex chronic conditions can't predict symptom flares. Crashes arrive without warning, 24–72h after the trigger.

N=1 longitudinal study: 207 nights of nocturnal HRV from a consumer wearable. Five independent models, each selecting its own features via forward selection across 13 candidates. Validated on 60 prospective pairs with LOO-CV.

AUC 0.84 (severity), 48h predictive lag from autonomic signals — not yet replicated in literature. All code and data public.

LLMs in clinical contexts need guardrails that aren't just prompt tricks. Prompt-based safety fails silently.

Five domain-independent axioms (Conciencia, Claridad, Límite, Pragmatismo, Cuidado). Deterministic evaluation pipeline: regex patterns, cosine similarity (0.92 threshold), gray zone flagging (≥0.75). No LLM in the evaluation path.

Every output evaluated before reaching the patient. APPROVE / REWRITE / SILENCE decisions with full audit trail. Three structural bugs publicly documented.