Production first
Every tool is designed to run in production from day one. No toy demos, no "you'll figure out the hard parts yourself."
Open source · AWS-native · Production-grade
Infrastructure for
AI engineering.
Tools and patterns for teams building real AI systems on AWS. From MCP servers to agentic orchestration — opinionated, documented, deployable.
$ pip install harrier-emr-mcp
Successfully installed harrier-emr-mcp-1.2.0
$ harrier diagnose --job-run <id>
PASS
Observability — log evidence available ISSUE
Spark Runtime — shuffle spill detected NOT CHECKED
Infrastructure — awaiting follow-up → Likely path: SHUFFLE_SPILL · Confidence: Medium-High
Products
Purpose-built tools,
not general frameworks.
Each product solves a specific problem in the AI engineering stack. Open source, AWS-native, designed for production.
Live ↗
Harrier EMR MCP
Evidence-backed initial diagnosis for EMR on EC2, EMR Serverless, and EMR on EKS failures. Built for AWS DevOps Agent.
In development
Merlin AI Ops
Agentic operations platform for AWS-native AI engineering teams. Bedrock AgentCore Runtime, Step Functions orchestration, multi-engine task execution.
What we believe
Platform engineering is a discipline, not a toolchain swap.
Bounded scope
Each product does one thing well. We do not build general-purpose frameworks. We build the specific thing a specific team needs.
Open and auditable
Source code is public. Decisions are documented in ADRs. You can read exactly how and why things work before you deploy them.
From the blog
Latest writing.
I tested Fable 5 on a real engineering project. Here is what happened.
A real-world stress test of Fable 5 on an active, mid-size project — 32k lines of code, AWS serverless, agents, Terraform. What worked, what didn't, and where long-running AI coding actually earns its cost.
Managing your network CIDRs across multiple AWS accounts in a Control Tower setup
How to use AWS IP Address Manager (IPAM) in a Control Tower environment to centrally plan, allocate, share, and monitor VPC CIDRs across accounts, regions, and AZs — manually and with Terraform.
Training
Learn by doing.
beginner
Getting started with Harrier EMR MCP
Install Harrier, connect it to your MCP client, and run your first EMR diagnosis. This module walks through installation, configuration, and a live demo run against an EMR Serverless job.
⏱ 30 min HarrierEMR
intermediate
MCP server design patterns for AWS operations
How to design MCP servers that give AI agents safe, bounded access to operational context. Covers tool design, evidence collection, pattern classification, and secrets handling — with examples from Harrier's source.
⏱ 45 min MCPAWS
Start building
Ready to run Harrier?
Install in minutes. Works with AWS DevOps Agent and any MCP-compatible client.