LifeLine

The Problem & Real-World Impact

In emergency healthcare, families lose critical hours manually broadcasting blood requests across WhatsApp groups with no coordination, leading to duplicate outreach, fake leads, and double-booked donors.

LifeLine solves this by combining natural-language AI parsing, sub-100ms geospatial donor search, and single-threaded Redis atomic distributed locking to guarantee zero double-booking race conditions under high concurrency.

Key Technical Architecture Highlights

• Atomic Distributed Locking (SET NX PX 900000): Single-threaded Redis lock acquisition eliminates double-booking race conditions when multiple requesters attempt to claim the same donor simultaneously. Validated under 20 concurrent requests — exactly 1 acquired the lock and 19 failed with a 409 Conflict.
• Polyglot Persistence Architecture: MongoDB Atlas $geoNear aggregation pipeline resolves top-10 compatible donors within 50 km in under 100 ms; Redis for sub-millisecond locks and instant session revocation; PostgreSQL + Prisma ORM for immutable audit logging.
• Advisory AI with Resilient Fallbacks: Natural-language request parsing uses OpenRouter (GPT-4o-mini) wrapped in an 8-second AbortController timeout and deterministic regex fallback (< 2 ms), guaranteeing the core application never hangs during medical emergencies.
• Dual-Token Auth & Instant Socket Revocation: 15-minute in-memory JWT access tokens paired with 7-day httpOnly, SameSite=Strict refresh cookies rotated in Redis. Logout triggers a WebSocket session-revoked broadcast that immediately terminates active sessions across all devices.
• Auto-Escalation State Machine: If a reserved donor fails to respond within the 15-minute lock window (900s TTL), the state machine applies a -10 reliability penalty, releases the Redis lock, and automatically re-matches the request with the next nearest candidate.

Live Concurrency Pipeline & Mutex Visualizer

Watch 20 concurrent hospital threads race to acquire an atomic distributed lock (SET NX PX 900000) for donor don_8f91a in real time:

Verified System Benchmarks

Tech Stack Breakdown