Raj Murugan
I build production AI agents on AWS
Deep technical posts with real code, real gotchas, real architecture decisions. No demo-grade content.
AWS Solutions Architect & AI Engineer · Sydney · Parallo (SoftwareOne)
19 technical posts · AWS Solutions Architect Professional · AWS Certified AI Practitioner
New: Your golden dataset is too easy Read →
Start Here
All posts →The AI Architect Roadmap
Eight rungs from foundations to enterprise AI, and an honest map of what this site covers deeply, where it is strong, and where the gaps still are.
Recent Posts
View all →Your golden dataset is too easy
I could not detect a deleted guardrail with an LLM judge, a judge-free assertion, or six label-free signals. Three instruments, one null. The instrument was never the problem: shorten the source and the same gate goes from p = 1.000 to p = 0.0020.
Your regression gate needs a power calculation
I deliberately broke my summariser's prompt, then failed to detect it two ways: with an LLM judge over a golden dataset, and with a judge-free deterministic assertion. Removing the judge changed nothing. Here is the calculation that would have told me first.
A clean pass rate is not calibration
I built an LLM-as-judge eval on my own blog and got a suspiciously perfect 16/16. Here's the three-round test I ran before trusting that number: single-variable corruption, and a self-consistency check the research says most teams skip.
About
Senior Solutions Architect at Parallo (SoftwareOne, AWS Advanced Partner). I take customers from "we want AWS" or "we want GenAI" to a working, governed deployment, then write about how it actually works and where it breaks.
The work I publish here is the work nobody else writes about: the IAM trust policy that takes an afternoon to debug, the VPC cold start that breaks streaming, the OIDC flow that silently fails for a month. AgentCore production gotchas. CDK patterns that survive a real deployment. Cost decisions that show up on the bill, not in the demo.
Featured Work
Production-ready Customer Service AI Agent on Amazon Bedrock AgentCore. Every deployment gotcha documented inline. The companion repo to the 6-part AgentCore series.
An MCP server that turns ADRs, incident reports, and runbooks into a queryable org-knowledge surface for AWS DevOps Agent. Four tools, Bedrock Knowledge Base, frontmatter-filtered chunks. From "agent that reads your docs" to "agent that knows your org."
Three Claude projects with Socratic tutoring, a weekly Cowork routine, three differentiated emails. Open-source spec, full design history including the v1-to-v2 architecture pivot. A personal build whose failure modes mirrored enterprise AI patterns.
Newsletter
Deep AWS + AI engineering in your inbox
A new write-up every Tuesday: the kind of post that documents an IAM gotcha, a real CDK pattern, or a production Bedrock deployment decision. No filler. No AI-generated content. Just one post a week worth your attention.
No spam. Every email has a one-click unsubscribe link.