Skip to content

Work

Systems and tools we have built

Three kinds: systems we operate, open-source software we publish, and AI models running on our own machines.

Systems

Systems we operate ourselves: store and back office, a development platform, support bots, deployment and operations. Open one for more screenshots and how it was built.

  • eSIM 226 storefront

    E-commerce

    eSIM 226 storefront

    A travel eSIM store in eight languages with multi-currency pricing and two Taiwanese payment gateways. Paid orders are provisioned with the upstream supplier automatically and the install email goes out on its own.

    • Eight locales and multiple display currencies, so Japanese buyers see yen
    • Automatic provisioning across two suppliers, with failover
    • Back office for plans, pricing, orders, refunds and live GA4 traffic
    • A LINE bot that answers plan and order questions directly

    Next.js · PostgreSQL · Keycloak · ECPay/NewebPay · Kubernetes

    See insideE-commerce siteVisit site
  • eSIM back office

    Admin system

    eSIM back office

    The operations side of the same store: products, pricing, orders, refunds, suppliers and payments in one place. It actively surfaces problems like "paid but never fulfilled" instead of waiting to be asked.

    • Orders, revenue, unpaid and unfulfilled counts at a glance
    • AI pricing suggestions and bulk listing changes
    • Refund review, low-data alerts and a report of searches that found nothing
    • Roles, permissions, audit trail and a login allow-list

    Next.js · PostgreSQL · Keycloak · GA4 Data API

    See insideE-commerce site
  • v-terminal developer platform

    Harness / internal tools

    v-terminal developer platform

    Our own browser-based platform for building and operating software: terminals, Kubernetes, Helm GitOps deploys, CI, registry and an AI coding agent in one screen. This website was shipped from zero using it.

    • Git push to self-hosted runner to Harbor to Helm sync, one pipeline
    • AI console: send a task from your phone, the agent reads the spec, edits and tests
    • Resources, events, logs and AI diagnosis without opening kubectl
    • An MCP server so other AI tools can drive the platform

    Next.js · Node.js · SQLite · Kubernetes client · Claude Agent SDK

    See insideHarness
  • AgentHub support and CRM

    LINE bot support

    AgentHub support and CRM

    Multi-channel AI support: LINE, a website widget and email land in one inbox. The bot answers first and hands over to a human when it cannot. Flows are drawn, not coded.

    • LINE, web widget and email in a single conversation thread
    • Drag-and-drop flow editor, so changing a flow is not a deploy
    • The bot calls real APIs to look up orders and plans
    • Voice message transcription, multi-tenant, role-based access

    Next.js · LINE Messaging API · PostgreSQL · Claude/Gemini

    See insideLINE bot support
  • AI video studio

    AI applications

    AI video studio

    Turns a PDF or a paragraph into a short video with narration, subtitles and motion: AI writes the script, generates the images and the voice, Remotion renders the file.

    • PDF or text to storyboard to narration to finished file, unattended
    • Narration and subtitles in several languages, with proper CJK line breaking
    • MCP tools, so another system can order a video in one sentence

    Remotion · Vertex AI · Qwen TTS · MinIO

    See insideAI applicationsVisit site
  • Virtual human and AI media generation platform

    On-prem models

    Virtual human and AI media generation platform

    A generation platform on our own GPU server. Each virtual human has its own LoRA, so the face stays the same across scenes and outfits; they shoot everyday clips and act in multi-character sketches, lip-synced in their own voices. It also turns a topic, a news article or a slide deck into a narrated, subtitled video, and can produce a whole instrumental album. Images, video, voice and music are all generated on the local GPU.

    • Virtual humans: one LoRA per character, the same face across scenes and outfits, each with its own voice and persona
    • Clips and sketches: AI writes the script, plans the storyboard, generates frames and shoots the video, and you can stop to review, edit or redo any step
    • Workbench: text-to-image, image-to-image, retouching, outfit swaps, image-to-video, lip sync, music and voice, with switchable models
    • The GPU queue runs one shot at a time and shows VRAM live; scripts can be written by Claude or by a local llama.cpp model

    ComfyUI · Z-Image · Krea 2 · Wan 2.2 · LTX · ACE-Step · local GPU

    See insideOn-prem AI model solutions
  • Voice training and fine-tuning

    On-prem models

    Voice training and fine-tuning

    Fine-tunes a dedicated voice from your recordings on a local GPU. A trained voice turns text into speech, with controls for stability, speed and emotion; a pronunciation table corrects Taiwanese Mandarin readings.

    • Each voice gets its own adapter; every epoch is kept, so you can play the same line on each version and pick one
    • Zero-shot cloning for when there is not enough audio yet: 3–15 seconds of clean speech gives a first preview
    • Pronunciation table for Taiwanese Mandarin: whole-word replacements and polyphone overrides, applied on save
    • HTTP API and MCP tools; AgentHub synthesises its LINE voice broadcasts with it

    Qwen3-TTS · CosyVoice3 · LoRA · local GPU · MCP

    See insideOn-prem AI model solutionsVisit site
  • Token Monitor team AI usage dashboard

    Open-source customisation

    Token Monitor team AI usage dashboard

    See how your team actually uses AI: who uses which tools, how much, at what cost, and how much of each subscription is left. We collect usage records from each computer with the open-source Token Monitor (MIT; it supports 43+ AI coding tools such as Claude Code, Codex and Cursor), combine the machines through a Hub, and add the web dashboard the project does not ship, so the whole team's usage is visible in a browser.

    • Per computer: which AI tools each colleague's machine uses, today's and this month's tokens and cost, and when it last reported
    • Reads only the usage records each tool leaves on the computer (token counts, model, time); it does not read prompts or code
    • Remaining quota and reset times for Claude, Codex and other plans, plus a daily usage trend for the last 14 days
    • We can also change Token Monitor's source (MIT licensed): store usage in a database, have new computers join the Hub automatically and run your own update channel, so rolling it out to a team does not mean setting up each machine by hand

    Token Monitor (MIT) · Hub API · Live updates · Web dashboard

    See insideAI token usage monitoring
  • Tech Tactician engineering blog

    Content site · AI automation

    Tech Tactician engineering blog

    Over 253 technical articles on AI, .NET Core, back end and DevOps. Articles go through an automated pipeline: research, outline and SEO analysis, writing, and a quality review that sends drafts back for a rewrite when they fail. After saving, the cover image, podcast and video are produced in parallel, and the video is uploaded to YouTube.

    • Over 253 articles with categories, tags, site search and RSS
    • AI writing pipeline: research → outline and SEO analysis → writing → quality review, rewritten automatically if it fails
    • After saving, cover image, podcast audio and video are produced in parallel; the video is uploaded to YouTube
    • Each step can switch model or be skipped; multilingual, dark mode, SEO structure handled automatically

    Next.js · MDX · Claude · Gemini · Veo · VoAI TTS

    See insideAI automation applicationsVisit site

  • AI automation

    Taiwan stock analysis and alerts

    Every day it pulls market data, computes indicators, writes an AI forecast report and a podcast episode, then pushes it to subscribers over LINE. Nobody presses a button.

    • Scheduled fetch, analysis, report and push, end to end
    • AI-written daily reports and a generated audio podcast
    • LINE account binding, subscribers choose what they follow

    Next.js · Python · LINE Messaging API · Vertex AI

    See insideAI automation applications
  • Low-code CMS

    Open-source customisation

    Low-code CMS

    A Strapi-style content platform taken a step further: a page editor, form backend, SEO settings and theme injection. Visitors can spin up a one-hour trial instance with one click.

    • Drag-and-drop page editor and a widget builder
    • Forms, SEO and multi-language content built in
    • On-demand trial instances, reclaimed automatically after an hour

    Next.js · PostgreSQL · Helm · Kubernetes

    See insideOpen-source customisation

  • Consulting / FDE

    Kubernetes, simplified: our deployment platform

    Kubernetes wrapped into a platform where pushing code is the deployment. Builds, the image registry, Helm releases and SSO are already wired up; a new service is one chart folder, and developers never touch kubectl. More than thirty of our own services run on it, and client projects use the same setup, in your own cloud account or on a cloud server bought on your behalf.

    • Push to main and it ships: self-hosted GitHub Actions runners to Harbor to Helm GitOps
    • A new service is one chart folder, synced by v-terminal; no hand-typed kubectl
    • SSO, mail and ingress pre-wired: Keycloak, Stalwart, Cloudflare plus frp
    • More than thirty services in long-term operation; outages, disk pressure and certificate rotation all handled

    RKE2 · Helm · Harbor · Keycloak · Cloudflare · frp

    See insideConsulting / FDE