The Hospital API faced critical performance bottlenecks, where slow retrieval of patient data was actively affecting care coordination efficiency in urgent scenarios. I spearheaded the optimization of the backend service, introducing advanced filtering mechanisms and database query optimization.
BACKEND & DATA OPTIMIZATION
Hospital API Patient Lookup Optimization.
The Hospital API faced critical performance bottlenecks, where slow retrieval of patient data was actively affecting care coordination efficiency in urgent scenarios. I spearheaded the optimization of the backend service, introducing advanced filtering mechanisms and database query optimization.
EXECUTIVE SUMMARY
KEY CONTRIBUTION
"Implemented a compound indexing strategy combined with Redis-based query caching that brought P99 response times for complex patient history lookups down from 3 seconds to under 200ms."
API ARCHITECTURE
High-level overview of the control and data plane components.
LOGIC LAYER
Query Optimization
Compound indexing & query generation
Advanced Formatting
Data transformation and hydration
DATA LAYER
Database
Main analytical patient records storage
Cache
Redis-based query caching layer
ENGINEERING STACK
High-reliability healthcare backend systems.
BACKEND
GoGingRPCGraphQL
DATABASE
PostgreSQLRedisElasticsearch
INFRASTRUCTURE
AWS EKSTerraformDocker
TECHNICAL DECISIONS
Documenting the trade-offs and architectural shifts during development.
Caching Strategy
Rationale: Patient health records are highly sensitive to staleness. Relying on TTL caching was deemed a clinical risk. We implemented a Postgres trigger system that emits an event to our Go service, actively invalidating the exact Redis keys affected by any write or update.
MEASURABLE IMPACT
Impact on clinical data retrieval speed.
QUERY LATENCY
200 ms
⚡ Down from 3000msDATA RETRIEVAL
60 %
⚡ Faster overall lookupPOSTMORTEM & LEARNINGS
Reflections on mission-critical medical systems.
Offloading the full-text search requirements from Postgres `LIKE` queries to a dedicated Elasticsearch cluster immediately stabilized our primary database CPU utilization, preventing cascading timeouts during peak operational hours.
Transitioning the monolithic Go service into domain-driven microservices (e.g., separating patient ingest from historical read models) using CQRS patterns to further isolate fault domains.