VitalSense: Real-Time Patient Triage on Redis (Streams • Search • JSON • TimeSeries)

This is a submission for the Redis AI Challenge: Beyond the Cache.

What I Built

A real-time patient triage dashboard that uses Redis 8 as the primary data platform. It ingests vitals (HR/SpO₂/BP) via Redis Streams, persists the latest state per patient in Hashes, indexes snapshots with RediSearch for instant querying, publishes alerts with Pub/Sub, stores adjustable alert thresholds in RedisJSON, and renders 24-hour trends from RedisTimeSeries. The React/Next.js UI is real-time (WebSocket), offline-first (IndexedDB hydration), supports threshold configuration (persisted in Redis), and provides a drill-down panel with a per-patient sparkline (via TS.RANGE).

Key properties:

• Primary DB: No external database; Redis is the system of record for runtime state + settings.
• Search & Rules: RediSearch powers numeric queries for “critical” patients (low SpO₂ / high HR), pulled by a watcher that publishes alerts.
• Streams + Pub/Sub: Streams for ingestion/fan-out; Pub/Sub for real-time alert fan-out to the browser.
• Time series: Per-patient SpO₂ time series with 24h retention for drill-down trends.
• UX: Progressive skeletons, dark mode, drill-down, offline cache, threshold modal (live-reconfigures the watcher).

Demo

Github:

devinnicholson / vitalsense

VitalSense Hackathon Project

A real-time health data pipeline using Redis Streams, RedisJSON, RediSearch Pub/Sub, and a beautiful Tailwind/Next.js dashboard.

---

Prerequisites

• Node.js v16+

• Python 3 (for the static dashboard, optional)

• Redis 8 instance with modules:

  • RedisJSON
  • Redis Streams
  • RediSearch
  • (Optional) RedisAI

• Environment variable

  export REDIS_URL="redis://default:<PASSWORD>@<HOST>:<PORT>"

---

1. Vitals Generator

Emits synthetic patient vitals into patients:stream.

cd vital-sense-generator
npm install
npm start

You’ll see logs like:

Emitted <streamId> { patientId: 'P01', hr: 82, spo2: 95, sys: 120, dia: 80 }

---

2. Stream Consumer

Reads from patients:stream, writes snapshots to patient:<ID>:snapshot, and acknowledges.

cd vital-sense-consumer
npm install
npm start

You’ll see:

✔ Consumer group "consumer-group" created on "patients:stream"
▶ Listening for new entries…
🗂 Got 1 msg(s) from stream "patients:stream"
✅ Processed <streamId> → patient:P01:snapshot { hr: '82', spo2: '95', sys: '120', dia: '80' }

---

3. Index Creation (RediSearch)

Drop & recreate the snapshots index to cover all…

View on GitHub

Demo:

vitalsense.mov - Google Drive

drive.google.com

How I Used Redis 8

Data ingestion (Streams)

• XADD patients:stream * id P01 heart_rate 82 spo2 95 bp_systolic 120 bp_diastolic 80
• Consumer group (XGROUP CREATE / XREADGROUP) provides reliable processing and back-pressure.

Primary state (Hashes)

• Latest snapshot per patient in Hash:
• HSET patient:P01:snapshot hr 82 spo2 95 sys 120 dia 80 timestamp <iso>

No external DB; Redis holds the live source of truth.

Indexed queries (RediSearch)

• FT.CREATE idx:snapshots ON HASH PREFIX 1 "patient:" SCHEMA hr NUMERIC SORTABLE spo2 NUMERIC SORTABLE sys NUMERIC dia NUMERIC timestamp TAG
• Query examples used by the watcher/UI:
• Low oxygen: @spo2:[-inf 92]
• High HR: @hr:[120 +inf]
• Combined: (@spo2:[-inf 92]) | (@hr:[120 +inf])
• Event fan-out (Pub/Sub)
• Watcher runs a periodic RediSearch query, publishes critical hits to alerts:critical.
• A Node WS proxy subscribes and forwards JSON to browser WebSockets.

Time series analytics (RedisTimeSeries)

Per-patient SpO₂:

• TS.CREATE ts:patient:Pxx:spo2 RETENTION 86400000 LABELS patient Pxx
• TS.ADD ts:patient:Pxx:spo2 <ts> <spo2>
• Drill-down: TS.RANGE ts:patient:Pxx:spo2 <now-24h> <now>
• Dynamic configuration (RedisJSON)

Alert thresholds live in JSON:

• JSON.SET settings:thresholds . '{"spo2":92,"hr":100}'
• Next API GET/PATCH /api/settings/thresholds reads/writes this key.
• Watcher reloads thresholds each loop → instant effect.

Team Submissions:
Devin Nicholson