Pravaah

The problem

Hospital care spans dozens of decisions across fragmented systems. Critical patterns get missed when nothing connects the dots — an O₂ saturation slipping 96%→91% over four hours triggers no alarm, but should.

How it works

Six specialist agents collaborate through Elasticsearch — Triage (MEWS scoring), Recovery (weighted recovery slope), Capacity (bed/ventilator optimization), Discharge (a conservative 7-point checklist), Guardian (deterioration detection, with veto power), and an Orchestrator coordinating all five. Its signature moment: Guardian catches hidden deterioration in a seemingly-stable asthma patient and blocks discharge.

Key features

• 6 collaborating agents including a Guardian with discharge-blocking veto power
• 13 ES|QL query tools + 4 automated workflows (decision logging, alerts, state updates)
• Time Series Data Streams holding 1,500+ vitals readings at 15-min intervals
• 8 synthetic patients with distinct clinical arcs that exercise each agent
• MEWS triage scoring and weighted recovery-trajectory computation

Architecture

A Kibana chat request hits the Orchestrator, which fans out to five specialist agents. Each uses ES|QL, workflow, search, and TSDS tools against a single Elasticsearch layer of five indices.