繁中 →
Leo Huang

Turn your SOP into an agent that actually works

Leo Huang · one-person company + an AI agent team I built myself

What I do day to day

I have a multi-agent system of my own, and I use it to operate a few live products: scheduling, pulling data together, producing content, watching customer support, monitoring service health. Most of the time it runs without me watching it.

You want to turn your team's SOPs, processes and decision trees into agents that really work, which is close to what I do every day. So what I hand you is something already running, with people using it, not just a plan.

The five directions you listed, I have built the equivalent

Here are the five agents from your post, matched against what I have actually built and still have running:

Sales Agent
My matching platform NaLi Match runs a flow for automatic matching, quoting and notifications: a request comes in, the right person gets found, a notification goes out, and the follow-up gets tracked. Organizing client information, generating quotes and pre-meeting briefs are very close in nature.
PM Agent
The system has a resident scheduling agent that runs tasks on a fixed rhythm every day, produces a morning report, watches progress, and writes up yesterday so the next day picks up cleanly. Project tracking, risk alerts and meeting notes are this same category.
Marketing Agent
I have a set of content agents that draft posts every day, take reference material apart, lay it out as images, and after publishing go back and confirm it actually went out. Content generation, campaign planning and asset management all live in there.
Finance Agent
For the platform's operating numbers, I have an agent query the database directly, compute the metrics and produce the report. Numbers get calculated by code rather than guessed by AI, and it all gets collected once a day. Financial analysis and reporting are the same job.
HR / support agent
I have also built an AI support agent on a LINE official account where customer messages get answered automatically and common questions get resolved on their own, 24 hours a day with nothing missed. Recruiting Q&A, onboarding and policy questions run on the same foundation.

I use all of these in practice, so they are the ground I know best.

Go take a look

These are all live right now, so you can open them directly:

NaLi Match is a nail and lash artist matching platform I built on my own with my AI team. It has over five hundred verified artists today, with real requests and real bookings happening every day, and matching, notifications, support and the back office are mostly automated. claude-real-video came out of a blocker I hit in my own work: an open source tool that lets AI read what is inside a video. One command installs it, and people outside Taiwan use it too.

The tools I use

Claude Code codex MCP Agent Workflow / multi-agent orchestration n8n Supabase Vercel LINE / Telegram Bot

I work with these every day. When I build an agent I habitually add checkpoints: have a second agent review it, test it for real before it goes live, compute numbers in code against the database. Agents built that way hold up better, and that is the only reason I am comfortable handing daily operations to them.

How we can work together

This can be flexible, freelance or project based. If it works for you, I suggest starting with the one process you most want automated, turning it into a single agent and getting it running. Once you see what it does, you decide whether to carry on with the rest. That keeps your commitment small and confirms the thing really works first.

Let's talk for 30 minutes

Tell me the one thing you most want automated first, and I'll walk you through how I would build it

cortexos.main@gmail.com