Who

I'm Jie Xiang. I work on reliability for a commercial KVM cloud platform — six years of testing, debugging and chasing long-tail bugs through compute, storage, networking, HA, live migration, the whole stack. The kind of platform that runs for enterprise customers where an incident means someone on a phone at 3 AM.

What I ship on the side

I run a multi-agent Claude Code pipeline that has put 32 iOS apps on the App Store as a solo dev. The orchestrator dispatches to a dozen specialized subagents — market research, UI design, Swift implementation, StoreKit subscription work, screenshot automation, App Review compliance, ASO. The portfolio lives at frexx.top; the Mac App Store companion at prereviews.app.

Two byproducts of that work are MIT-licensed:

• apple-presubmit-audit — 70+ App Store Review checks built from ~50 real rejections. Catches privacy manifest misses, missing "auto-renewing" CTAs, declared-but-unused permissions, and more before you hit Submit.
• agent-ledger — per-subagent cost attribution for Claude Code. Reads the local JSONL session logs and shows what each subagent actually consumed at marginal API rates (the "shadow cost").

What I'm building now

Pointing the same reliability rigor at AI Agents. I'm one of the earlier builders in the Claude Code and Model Context Protocol (MCP) ecosystem, writing tools around the parts that are still rough — cost attribution, debugging, reliability primitives. This site is where I publish what I find.

Working together

I'm open to a narrow set of engagements:

• AI Agent evaluation and observability — for teams shipping agents to production
• MCP server development — custom integrations for infrastructure and internal tools
• Cloud infrastructure reliability consulting — KVM, virtualization, hypervisor tuning

Reach me: jie.xiang.jm@gmail.com · Bluesky · GitHub

Reading order if you're new

The best entry point is the most recent post on the blog. Everything here is written for engineers — no tutorials, no hype, no "leveraging cutting-edge synergies." If a post doesn't have a concrete number, a concrete bug, or a concrete decision in it, I don't ship it.