I help small teams turn repetitive work into AI-assisted systems with logs, review gates, cost awareness, and clear operating docs.
They need one repetitive workflow cleaned up first.
Before building an agent, I look for the smallest repeatable system that can safely reduce work. Sometimes the right answer is automation. Sometimes it is cleanup first. Sometimes it should stay manual.
Not generic bots. Small supervised systems around real recurring work.
Scoped around input, output, review, logging, cost, and operating docs.
Scoped around input, output, review, logging, cost, and operating docs.
Scoped around input, output, review, logging, cost, and operating docs.
Scoped around input, output, review, logging, cost, and operating docs.
Scoped around input, output, review, logging, cost, and operating docs.
Scoped around input, output, review, logging, cost, and operating docs.
Scoped around input, output, review, logging, cost, and operating docs.
Scoped around input, output, review, logging, cost, and operating docs.
Start where the task is repetitive, expensive enough, and easy to review.
The default path is async. Submit one workflow first; calls happen only when needed.
Send the workflow, sample input/output, and current pain point. I review it and send back a short assessment: automate, clean up manually first, or avoid AI.
One workflow, one working prototype, logs or review gate, operating doc, and async handoff.
A deeper 2-4 week implementation for teams that need a more reliable internal workflow.
Ongoing workflow improvement, prompt/agent updates, quality checks, and small additions.
For important work, an AI system should be inspectable and interruptible.
Built into the operating flow so the team knows when to trust the output and when to stop.
Built into the operating flow so the team knows when to trust the output and when to stop.
Built into the operating flow so the team knows when to trust the output and when to stop.
Built into the operating flow so the team knows when to trust the output and when to stop.
Built into the operating flow so the team knows when to trust the output and when to stop.
Built into the operating flow so the team knows when to trust the output and when to stop.
Yes, when a chatbot is actually the right interface. But I usually start with the workflow: input, output, review step, logs, and fallback path. Many teams do not need a chatbot first.
Usually yes. The default flow is async: you submit the workflow, I review it, then I send a quote or audit recommendation. A short call happens only if needed.
Recurring coding, data, research, reporting, document review, content ops, and internal operations tasks with clear inputs and outputs.
Vague “use AI somewhere” requests, fully autonomous sensitive decisions, tasks with no clear owner, and workflows where the desired output cannot be described.
Not perfect data. But the mess needs to be visible. If the input is too inconsistent, the first step may be manual cleanup or a data QA workflow.
Only with extra scoping, anonymized samples first, and clear security boundaries. Do not submit credentials, private customer data, or secrets in the quote form.
AI Agent Pilots start from 12M VND / $700. Smaller audits start from 3M VND / $150. Larger workflow sprints start from 35M VND / $2,000.
Yes. I can work in English or Vietnamese. The website is English-first for global reach, with Vietnamese intake templates for local teams.
I classify the workflow as worth automating, needs manual cleanup first, or not worth automating yet. If it fits, I send a quote range or recommend a paid audit/pilot.
I’ll tell you if it is worth automating, needs manual cleanup first, or should stay manual for now.