VitalGuard.ai

Signin/signup
initial version of stability in patient
Features
Calendly

Inspiration

Most existing patient monitoring systems react only after vitals cross dangerous thresholds. We kept asking a simple question:

What if a system could warn families and doctors before a medical emergency actually happens?

VitalGuard was built to shift healthcare monitoring from reactive alerts to predictive, autonomous decision-support.

Our goal was to design a system that not only detects deterioration early, but also coordinates response automatically — even in low-connectivity environments.

What it does

VitalGuard is an AI-powered real-time patient monitoring system that predicts medical deterioration using wearable vitals and autonomously coordinates emergency response workflows.

It ingests vitals from:

Bluetooth wearable devices (HR, HRV, SpO₂, temperature)
GalaxyPPG real signal streams
Synthetic simulator scenarios for testing

Each reading flows through a LangGraph agent pipeline that:

computes risk scores (0–100 severity index)
predicts deterioration using trend analysis
retrieves 7-day patient memory
generates AI clinical reasoning
provides personalized first-aid guidance
dispatches multi-channel alerts
discovers nearby doctors automatically
escalates if the patient does not respond
enables offline diagnosis using local medical LLMs

How we built it

VitalGuard is built as a state-machine driven agentic healthcare pipeline using:

LangGraph for orchestration
BioMistral for clinical reasoning
MedLlama2 (Ollama) for offline diagnostics
Firebase Firestore for longitudinal medical memory
Web Bluetooth GATT for wearable vitals ingestion
SerpApi for nearby doctor discovery
Twilio, Telegram, Email for escalation alerts
n8n for autonomous recovery workflows
Calendly for consultation scheduling

Vitals flow through a 12-node agent pipeline:

ingest → risk → trend prediction → memory retrieval → reasoning → first aid → escalation → triage → doctor discovery → persistence

Key innovations

VitalGuard introduces five major agentic capabilities:

Predictive deterioration detection Trend analysis predicts CRITICAL conditions before thresholds are crossed.

Longitudinal medical memory AI reasons using 7-day patient history instead of single readings.

Autonomous triage coordination Alerts, reasoning, and first-aid instructions execute in parallel.

Recovery workflow automation If alerts go unacknowledged, consultations are auto-booked and voice calls triggered.

Offline diagnostic intelligence MedLlama2 enables differential diagnosis without internet connectivity.

Challenges we ran into

Integrating real-time wearable vitals with an autonomous agent pipeline during a hackathon environment was one of the biggest challenges.

Designing a system that works both online and offline required building fallback reasoning layers using local LLM inference.

Coordinating multiple escalation channels while maintaining deterministic routing inside the LangGraph pipeline also required careful architecture design.