José Enrique Núñez García

Led the migration of a 250M+ item embedded KV store onto Redis + AWS DocumentDB with zero downtime, unblocking 3x traffic growth. The zero downtime side relied on dual writes with cleanup on partial failure and CAS semantics keeping legacy clients working through the cutover.

Optimized the request hot path across Go and Node.js services running 2M+ req/s (peaks 8M/s) under sub-50ms budgets, introduced zstd compression between services, restructured hot-lookup data structures, and cut syscalls/I/O. Removed ~500µs (~1 % of the latency budget) of system health overhead from the critical path.

Lowered per request CPU on the busiest services: data lookups 2-3x faster and filtering up to ~9x faster on the heaviest workloads, freeing tens of cores at sustained load.

Redesigned the shared cache client used across every backend service to support multi-node sharding via consistent hashing, then ported the same model into the Go service on the request hot path. Unblocked horizontal scaling of the cache without any migration work for downstream consumers.

Built the canary deployment system, replacing an all at once deploy model where any regression meant a full fleet rollback. Includes the instance manager, a watchdog, the deploy/promote CLIs, and the CI/CD orchestration around them. Powers releases of the busiest services across 4 AWS regions.

Built the real-time tuning engine that adjusts bid ratios against each tenant's gross margin target from streaming performance feedback. Consumed by the platform on every request, replicated across regions for global consistency, with safety guards (rate limited adjustment steps, bounded ratios) on every update.

Hardened and optimized a production real-time ML classifier (random forest, Go) running on the request hot path with continuous retraining. Added class balancing on the training samples to fix the positive/negative skew, stabilized the training pipeline, and wired the feedback loop that turns downstream outcomes into new training samples. Optimized model synchronization between instances, cutting traffic by ~17x per instance (from one request every 11.5ms to one every 200ms).

Built the compliance enforcement layer in the request hot path. Handles policy checks per region, newer consent framework versions as they roll out, vendor disclosure validation, and consent propagation through to downstream third party calls. Keeps EU traffic GDPR-compliant at 2M+ req/s, no measurable latency hit.

Designed and developed an ecommerce application using Java Spring and Angular during my internship.