Aswitha Jadda

Scale and maintain multi-region distributed model registries using Go and Python, implementing custom connection pooling and gRPC health checks to support 45,000+ concurrent queries under a 99.99% uptime SLA. Design and deploy multi-tenant model-serving infrastructure on distributed AWS GPU clusters, targeting p99 latency under 120ms for serverless inference APIs during unpredictable peak traffic. Optimize training data streaming by building an Apache Arrow/Parquet chunk loader that reduces runtime memory consumption by 34%. Improve preprocessing performance with high-speed Rust tokenization routines (42% reduction in preprocessing latency). Architect distributed multi-agent memory abstractions with LangChain/LangGraph to keep cross-host context synchronization overhead under 15ms, overhaul caching with Redis and Pinecone to reduce cold-start delivery latency by 40% (85ms absolute speedup), and productionize post-training distillation pipelines using TRL to compress foundational LLMs i

Decomposed legacy monolithic components into decoupled Spring Boot microservices, cutting CI/CD deployment time from 45 minutes to under 12 minutes (73% reduction). Redesigned multi-tenant REST APIs with strict validation layers to handle 8,000+ RPS while reducing p95 response latency by 18%. Re-architected an event-driven data sync pipeline using Apache Kafka, scaling to 15M+ daily payloads with deterministic delivery constraints and an end-to-end message drop rate under 0.01%. Implemented multi-tier distributed caching with Redis clusters to achieve an 82% cache hit rate and reduce primary MySQL CPU spikes by 25%. Added resilience patterns (Resilience4j circuit breakers and fallbacks) to maintain cross-service availability SLA of 99.95%.