Built a highly available, event-driven microservice architecture in Go to orchestrate autonomous AI workloads at scale. Durable Execution: Engineered a zero-data-loss pipeline using Temporal.io to guarantee state resurrection if a worker node crashes during multi-step LLM inference tasks. Semantic Caching: Integrated Qdrant Vector DB via gRPC to build a sub-15ms semantic interceptor, caching mathematically similar prompts at the edge to completely bypass redundant AI API compute costs. Event-Driven Decoupling: Separated client ingestion from execution using Apache Kafka for high-throughput message brokering and Redis for distributed L7 rate-limiting. Real-Time Telemetry: Eliminated HTTP polling bottlenecks by implementing Redis Pub/Sub combined with Server-Sent Events (SSE) for full-duplex, real-time client communication.