---
id: cupix
type: experience
title: Software Engineer @ Cupix
url: https://auejin.com/en/experience/cupix/
lang: en
alternate:
  ko: https://auejin.com/ko/experience/cupix/
updated: '2026-08-11'
---

# Software Engineer @ Cupix

May 2025 – Present

- **What** — A company that turns buildings and construction sites into 3D digital twins people can look back through. I work on the frontend and as technical PM.
- **Why** — Sites lose network, yet the desktop app fetched everything from the server, and QA filed bugs faster than the team could absorb them.
- **How** — I built an offline mode that stores every entity's metadata and every image locally, and a hybrid layer that switches between the real API and the local database according to connectivity without changing a single call site. Edits made offline synchronise automatically on reconnection.
- **New** — I proposed and ran the company's first workflow in which AI agents carry out development work, and served as technical PM for Beacon, its conversational AI product.

## Product — SiteView Desktop

An Angular-based Tauri v2 desktop application.

- **Offline mode**: designed and implemented LocalDB and LocalStorage interfaces that store every
  entity's metadata and image assets locally, so the UI reads everything it needs without a
  network. Demonstrated to Korea Power Engineering Company at a quarterly progress review.
- **Hybrid mode**: designed a structure that calls the real API or emulates it from LocalDB
  depending on connectivity, without touching the engine libraries' call sites. Edits made
  offline sync automatically once the connection returns.
- **Connect**: implemented GeoJSON export so SiteView data opens in ArcGIS and QGIS, and
  refactored panorama download to work in ZIP batches. Replaced the dialog-based download and
  upload flow with a wizard and consolidated the services behind it.

## Bringing AI into the workflow

- **[Issue Fixer](https://auejin.com/en/projects/issue-fixer/)**: built and deployed an agent framework
  that collects, analyses, and fixes QA bugs, plus a dashboard for live KPIs and failure analysis.
  After adding a workflow that reads its own biweekly report and rewrites its own code,
  root-cause accuracy went from 78.5% to 92.2% and wasted PRs from 68.4% to 53.4%.

- **[Beacon Chat](https://auejin.com/en/projects/beacon-chat/)**: under a tight deadline I designed an
  agent-orchestration harness and built the chat server on it — the company's first workflow for
  **building features that did not exist**, rather than repairing ones that did. I served as
  **technical PM** for the product, splitting the spec into stories, distributing work across the
  team, and automating the dependency graph so the current bottleneck was always unambiguous.
- **local-serve**: automated the build-target lookup and token injection that used to dominate bug
  reproduction. The frontend and QA teams use it daily.

## Team and infrastructure

- Led the design and adoption of an AI-then-Human workflow across Jira, Slack, GitHub, and Azure
  for the frontend team.
- Rebuilt CI/CD for the monorepo so Jira tags, PR labels, and commit logs are recorded
  automatically, leaving metadata the agents can analyse later.
- Owned GitHub Actions runner capacity for the frontend team, optimised the pipeline, and locked
  down dashboard and webhook access behind a TailScale HTTPS funnel.

## Related

- [Beacon Chat](https://auejin.com/en/projects/beacon-chat/)
- [Isopod](https://auejin.com/en/projects/isopod/)
- [Issue Fixer](https://auejin.com/en/projects/issue-fixer/)