Anhphong — Founder & CEO at DEMETIO

About

Anhphong founded DEMETIO in 2021 to bring Japan-market-grade Shopify engineering out of Vietnam. The team now operates as a certified Shopify Partner with 40+ delivered projects ranging from D2C apparel, food, and B2B industrial commerce to AI-integrated storefronts and AR-assisted workflow systems.

Before DEMETIO, Anhphong spent the better part of a decade inside Japanese startups, shipping product across web, commerce, and internal tooling. JLPT N1 fluency and direct experience inside a Japanese org chart means proposals account for the kaizen reviews, vendor handover ceremonies, and seasonal campaign cadences that Japan EC actually runs on — not just the technical spec.

What I write about

Shopify Liquid & Hydrogen: performance, theme architecture, headless storefront patterns, Oxygen deploy.
Custom Shopify apps: App Bridge, billing API, security boundaries, multi-tenant data design.
AI integration for commerce: recommendation engines, support chatbots, content generation pipelines.
Japan EC playbook: konbini / COD / Paidy payment paths, JP-style logistics, cross-border tax + duty considerations.
Vietnam-Japan offshore practice: contracts, transparency, handover, runbook quality bar.

Approach

The DEMETIO bar is plain: code that another engineer can pick up six months later without me in the room. Every project ships with a runbook, a documented deploy path, transparent sprint progress in Notion / Linear, and weekly demos. We do not subcontract to freelancers; the engineer you see in week one is the engineer who pushes the production commit.

Selected projects

A handful of projects I have personally led or contributed to as principal engineer, picked because they each stretched a different dimension of what Shopify can carry.

Lumber procurement system on Shopify (B2B)

Took a phone-driven B2B lumber business onto Shopify with custom cut-to-size order forms, customer-tier wholesale pricing via tag-gated price toggling, and split shipping across three warehouses. Built as a custom theme plus a focused Shopify app rather than a heavy ERP integration. The operations team cut order-entry time from 15 minutes per order to 3 minutes and reduced inbound order phone calls by 70% month-over-month. Case study →

AI × AR glasses work guidance (industrial)

Connected a Shopify-managed content backend to AR glasses worn by manufacturing-floor operators. LLM-generated step overlays adapt to what the operator sees through the lens. The content authoring side lives in Shopify admin so non-engineers can publish new procedures without involving the dev team. New operator onboarding time dropped 40%; senior operator shadow hours dropped by more than 100 hours per month. Case study →

AI product recommendations (apparel D2C)

Connected Shopify Storefront API to a personalization engine that blends browse history, purchase history, and seasonal signal. Recommendation CTR went 2.8x; AOV lifted 15%; repeat-purchase rate moved up. The bigger lesson was on engineering discipline: the recommendation engine sits behind a strict response-time budget so it never degrades the storefront when the upstream API is slow. Case study →

Q&A

Why Vietnam-based for Japan EC engineering?

Two practical reasons. First, the 2-hour timezone offset (GMT+7 vs JST GMT+9) makes real-time collaboration possible in Japanese business hours — the friction you get with US- or India-based offshore teams disappears. Second, Vietnam's developer talent pool is deep (~50K new IT graduates per year) and skilled in the React / Node.js / Liquid stack that Shopify Plus and Hydrogen demand. The trade is that you have to invest in Japanese-language fluency on your side; DEMETIO's hiring bar is JLPT N2 minimum to remove that friction from the client side.

How do you think about Liquid vs Hydrogen for new builds in 2026?

Default to Liquid until a specific reason to go headless appears. Liquid themes ship faster, are cheaper to maintain, and reach 95+ Lighthouse with discipline. Hydrogen earns its complexity premium when you need a sub-second TTFB on every route, deep customization beyond what Online Store 2.0 sections allow, or when the storefront is part of a larger React-based product (e.g. embedded inside a SaaS dashboard). Roughly one in five of our new builds goes Hydrogen — that ratio reflects when it actually pays back.

What does AI integration in Japan EC actually look like in 2026?

Three patterns dominate. (1) Personalized product recommendations connected via Storefront API — the biggest revenue lift per dollar of engineering. (2) Support chatbots in Japanese, usually GPT-4-class with retrieval grounding on the merchant's policy + product docs. (3) Content generation pipelines for product description seed text in JP/EN — useful for high-SKU catalogs, but every generated string still needs human editorial. The pattern I am consistently skeptical of: replacing the product page with a chatbot. Japanese D2C buyers want structured product info, not a conversation.

What is the most expensive mistake you see Japan-targeting EC merchants make?

Picking a payment stack that locks out konbini + COD. International merchants entering Japan often default to Stripe + Shop Pay, then discover they have ceded 20-40% of the addressable customer base. The fix is rarely technical — it is admitting at the planning stage that Japan EC needs konbini, COD, and ideally Paidy or atone, and budgeting integration time for them. The second-most-expensive mistake is going Hydrogen without a runbook for the deploy pipeline; we have taken over more than one Hydrogen project where the original team left and nobody knew how to push to Oxygen.

How do you handle the JP/EN bilingual delivery in practice?

Default working language is Japanese for client-facing communication, code review in English with Japanese commit messages allowed. Documentation is dual-language for anything client-touching (specs, runbooks, decision logs); internal engineering docs are English-only to keep them fast to write. Slack / Notion / Linear all in Japanese as the surface; meeting transcripts auto-translated for asynchronous review. The bilingual overhead is real, but the alternative — losing nuance in requirements gathering — costs more.

Where to find me

X / Twitter is the most active surface (linked above). For project enquiries, email or the contact form reaches me directly — every form submission lands in my inbox, not a sales queue. I try to reply inside one Japan business day.