Cloud-Native Microservices

P95 latency: 840ms → 120ms. Zero auth incidents after migration.

Decomposed a high-traffic monolith into domain-bounded Node.js services with REST + GraphQL. GraphQL schema stitching at the API gateway. JWT auth with Redis-backed token invalidation. Horizontal scaling on AWS ECS + GCP Cloud Run behind an ALB. P95 API latency dropped from 840ms to under 120ms.

a Node.js monolith with 200+ routes hitting a single Postgres instance at 50K concurrent users

Highlights

• GraphQL schema stitching at the gateway: each downstream service owns its SDL; the gateway merges them — frontend queries one endpoint, routing is invisible.
• JWT access tokens (15min TTL) + Redis-backed refresh token store with sliding expiry; token revocation is O(1) — Redis DEL, no DB round-trip.
• AWS ECS (Fargate) for stateless services; GCP Cloud Run for burst workloads — cost optimized by routing non-latency-critical jobs to Cloud Run spot instances.
• PostgreSQL read replicas for reporting queries; connection pooling via PgBouncer reduced DB connections from 3,200 to under 400 at peak.
• Structured JSON logging (Winston + CloudWatch Logs Insights) — queries that previously took 20min to triage now resolve in under 2min.

See full portfolio: https://portfolio.kunal-gautam-570.workers.dev