AURA: Ambient Wellness Companion

AURA (Ambient Wellness Companion) is a Kaggle Agents for Good capstone that runs ambient wellness check-ins for elderly patients. Built on Google's Agent Development Kit (ADK 2.0), the system orchestrates a state-driven multi-agent graph: a medication extractor maps patient language to schedule IDs, an empathetic companion agent scores mood and replies in natural language, a deterministic privacy guard persists only allowlisted telemetry fields, and a rule-based escalation engine routes critical cases to caregiver alerts.

The demo UI includes a passcode-gated patient dashboard with live JSON and agent activity, a provider view for roster management and alert triage, and an architecture overview page. Tool access follows least privilege via filtered MCP toolsets — LLM agents cannot write directly to the secure mock database; persistence happens only in the deterministic privacy node after structured output is available.

Highlights

• ADK 2.0 multi-agent workflow: medication extraction, empathetic companion, privacy guard, and escalation routing
• Deterministic persist_metrics_node with allowlist-validated writes (mood, compliance, per-med status)
• Escalation when consecutive missed medication cycles reach 2 or mood score drops below 3
• Patient, provider, and about views with passcode-gated sessions and live activity sidebar
• Filtered MCP toolsets enforcing least-privilege boundaries between agents and the secure store