System design, AWS, microservices, AI/ML deployment — practical posts for engineers who want to build production-grade systems.
Your agent's container says 'no internet', but can it resolve DNS? A practical guide to containing tool-using AI agents: deny-by-default egress, a controlled resolver, scoped credentials, and audit logs.
OpenAI pauses frontier training after an agent reached the internet through a DNS gap, Anthropic ships a faster and cheaper Sonnet 5.5, and xAI releases Grok 4.7 for long-running coding tasks.
What the nines actually mean, how to eliminate single points of failure, active-active vs active-passive redundancy, multi-region failover, and setting RPO/RTO targets that match the business, not just the architecture diagram.
Delivery guarantees, Kafka partitions and consumer groups, RabbitMQ exchanges, dead letter queues, and how to decide when async processing is the right call and when it just adds latency.
Treating prompts as versioned, tested, observable production code — prompt structure, few-shot examples, chain-of-thought, template reuse, compression for cost, and catching prompt regressions before users do.
How distributed tracing actually works — span propagation, correlation IDs, sampling strategy, and how to go from 'the API feels slow' to the exact service and query causing it.
Instrumenting a Spring Boot service end to end — OpenTelemetry setup, Prometheus metrics and alerting rules, Grafana dashboards, structured logs with correlation IDs, and how to design alerts that don't cause fatigue.
Database decisions outlive most code. This post covers relational vs. document vs. key-value vs. time-series vs. graph databases, polyglot persistence, sharding vs. replication, schema evolution, and the decision framework senior engineers use to pick the right store for the job.
How production systems survive dependency failures — circuit breaker states, timeout strategy, bulkhead isolation, retries with jitter, and fallback design, with real Resilience4j code.
Scaling RAG past a demo: choosing a vector database, chunking strategy, hybrid search and reranking, measuring retrieval quality, and where the cost actually goes at high query volume.