Xiaohui · engineer and writer
I build Agent systems for real work, and use code to expand my abilities and study investing. These are my projects, practice and ongoing ideas.
1M+ MAUInsurance Quick Check
SELECTED WORK / 01
A portfolio in progress
From model access and AI workflows to whiteboards, recording summaries and wireframes. A home for products, tools and ideas I keep building and sharing.
aispeeds.me
02 / BUSINESS PRACTICE
Agent engineering for an Ant Insurance product serving over a million monthly active users: product reports, quality scoring and evaluation workflows.
03 / OPEN SOURCE
An MCP tool for discovering and installing MCP servers, so useful tools are easier to find and connect to an Agent workflow.
2025 OceanBase AI Hackathon · Second Prize
IN PUBLIC / 02
AI MAKER SHANGHAI
Using Insurance Quick Check to explore how context, collaboration protocols and evaluation connect to real delivery.
Read the slides & notes · 中文A talk in the CCF Technology Forum on Agents and financial innovation.
Read the CCF event recap · 中文ONGOING QUESTIONS / 03
Build Agents that can do the work.
A place to begin
Expand your abilities through practice.
A place to begin
Let evidence test your judgment.
A place to begin
FIELD NOTES / 04
Engineering problems, observations from life, and investment decisions I revisit as the evidence changes.
Subscribe via RSSRecently updated first
How a report system combines scoped research tools, deterministic graph stages, safe artifact updates, quality gates, an…
A worked insurance-renewal example showing contrastive search, research traces, evidence boundaries, and evaluation for…
A Python Agent runtime from production practice, with shared LangGraph graphs, isolated runs, event-loop ownership, chec…
A 1-hour consultation with Xiaohui about AI Agent implementation, MCP, context engineering, AI Coding, and AI engineerin…
An initial English draft of a personal investment review on Kweichow Moutai, position sizing, and revised expectations.
Xiaohui is an AI Agent and MCP practitioner, Ant Group P7 engineer, lead of the Insurance Quick Check Agent, and author…
An initial English draft of a framework that evaluates assets by asking whether future cash flow comes from healthy debt…
An initial English draft of Xiaohui's OneAgent and context engineering methodology, inspired by Claude Code and Manus-st…
An initial English draft about turning valuation into position, execution rules, holding logic, and selling discipline.
An initial English draft about government as public-credit trustee, rule enforcer, risk backstop, and dangerous debt exp…
An initial English draft about the OneAgent + MCPs paradigm for creating reusable domain Agents inspired by Manus and Cl…
An initial English draft connecting the secretary problem to investment observation windows, position discipline, and ac…
An initial English draft about how ordinary people can use debt economics to choose cities, companies, careers, and majo…
An initial English draft about self-talk, repetition, identity scripts, and turning desired behavior into executable tri…
An initial English draft about travel as a way to see the deeper geography, industry, and living systems beneath standar…
An initial English draft of the argument that production Agent systems converge around stable protocol objects such as T…
A timeline of Xiaohui's path from study and engineering practice to AI Agent work.
XIAOHUI / KEEP IN TOUCH
More about my background, how I got here, and the projects I work on.
