Anatoly S.

12 Weeks to Staff Engineer: Week 3 Ship Log Week 3: The Knowledge Layer & Enterprise API Gateway This week, I transformed my multi-agent Swarm from a local terminal script into a deployable Enterprise Microservice, entirely powered by Anthropic's Claude. Built: - Migrated the LangGraph orchestration engine to Claude 4.6 Sonnet for superior parallel tool execution. - Implemented Decoupled RAG: The agent doesn't hold the Vector DB (ChromaDB). Instead, the DB sits behind the Model Context Protocol (MCP) server as a strictly gated tool. - Engineered "Null State" Fallbacks: If the Vector DB returns no internal policies, Claude is strictly prompted to execute a graceful fallback rather than hallucinating compliance rules. - Wrapped the entire LangGraph ecosystem in a FastAPI REST Gateway, heavily utilising async Python and Pydantic validation. The swarm can now be integrated into any front-end or banking application via HTTP. The Architect's Lesson: The hardest part of RAG isn't vector math; it's security. By putting ChromaDB behind an MCP server, prompt injection attacks against the LLM cannot arbitrarily drop database tables or read unauthorised documents. Blast Radius Isolation is critical. Follow the Build Log: https://lnkd.in/eq_VQYk4 #AgenticAI #FastAPI #Claude #LangGraph #StaffEngineer #MLOps #Python

2

What if you could easily understand how changes to your DB queries were influencing your end user experience no matter how complex the backend service landscape in between is? At Grafana Labs, we’ve always been about making complex things simpler. The first tool we built was Grafana, our single pane of glass.  From there we grew a forest of open source tools: Mimir, Loki, Tempo, Pyroscope, Faro and more — while also contributing to projects like Prometheus and OpenTelemetry. Using these open source tools we built open observability cloud solutions for gaining deep insights into your cloud providers, your infra components, your k8s clusters, your applications, and your frontends. Now we’re adding a solution for Database Observability (in public preview) to help you understand query performance, and even get AI-powered suggestions for improvement! And we’ve integrated the insights from all these solutions with the Grafana Knowledge Graph so you can understand how the components of your systems behave and interact! Simplifying observability doesn’t mean hiding complexity. It means giving you the tools to master it. The ObservabilityCON 2025 keynote is available on demand so you can catch the Knowledge Graph demo, and all the exciting announcements we shared: https://lnkd.in/g_bSPhCK

23

Running Kubernetes workloads on AWS and struggling to keep storage costs under control? K8s EKS clusters are often top contributors to redundant storage costs due to unattached EBS volumes. This commonly occurs when worker node volumes (used for ephemeral volumes) or EBS-backed PersistentVolumes aren’t configured for automatic deletion, leaving them behind after nodes or workloads are terminated. With the newly launched Automated Amazon EBS Volume Optimization feature in AWS Compute Optimizer, you can now apply organization-wide automation rules powered by automatically generated recommended actions. A single rule can operate across all Regions, removing the need to manage separate configurations per Region. Whether you’re cleaning up dev environments or optimising your production infrastructure, these automation rules bring the scalability, consistency, and safety needed to streamline your optimization efforts. https://lnkd.in/dqErX4An

9

Building Linux eBPF/XDP apps on macOS (Apple Silicon) I spend a lot of time building high-performance DNS filters and network tools in Go. My primary development machine is an Apple Silicon Mac, but all the production deployments target Linux/AMD64 servers. Recently, I started heavily integrating eBPF and XDP to drop abusive traffic (like random subdomain attacks) before it even hits the Go networking stack. However, compiling C eBPF code locally on macOS has always been a pain due to the missing Linux headers and the architectural differences... I wrote down the exact step-by-step process, including a workaround for the notorious Debian multiarch header issue (asm/types.h) and a neat little smoke test using bpftool inside the container..: https://lnkd.in/egqdQAd9

8

3 Comments

🚤 Brilliant Kubernetes Gateway API benchmarks from my colleague John Howard covering Envoy Gateway, Nginx, Istio Ingress (Envoy) and agentgateway. This is by far the most in-depth comparison I've seen covering scenarios like scale, performance, configuration propagation, failover scenarios and more in such great detail. Agentgateway is looking pretty good! See the full writeup linked in comments 👇

190

8 Comments

So the first time I saw 'Diskless Kafka' advertised at a conference I was suspicious. My mind went straight to 'Serverless' Compute... which despite the name seemed to require a whole lot of servers. Yes, I know, just not ones you manage yourself. But still. Well I've finally dug into 'Diskless' Kafka and it's actually well named! The brokers genuinely offload storage to object stores like S3, which changes the economics and operational overhead pretty dramatically. This article by Abhay Soni lays out the basics clearly with a nice comparison chart to boot. What catches my eye: tiered storage isn't just a cost play. It fundamentally changes how you think about retention and replay. That's a big deal. Curious what others are seeing in the wild with tiered storage adoption?? https://lnkd.in/eXJUdmJF

34

3 Comments