Lab Support

The problem

Requests between the lab, customer service and our office in Tokyo lived in email, Teams messages and hallway conversations. Nobody knew who owned a request, nothing had a deadline, and the lab manager had no way to see what was running late.

A colleague had prototyped a fix in SharePoint and Power Automate. It couldn't keep up: the form was rigid, the list didn't work as a queue, and there was no way to track deadlines or report on anything.

Discovery

I wrote a short proposal and a click-through presentation that answered the obvious questions before anyone had to ask them, then demoed a working build to the lab manager and two group leaders.

Her notes from that meeting became the plan: her seven categories, five priority levels, an audit trail on every ticket, and separate "assigned to me" and "I submitted" views, because the group leaders had found a single list confusing. I shipped it on her date, four days later.

Edge cases I caught

Her deadline rules mixed calendar time and business hours. Read literally, a High ticket would have had a later deadline than a Normal one. I raised it with her in writing and shipped with a written assumption instead of guessing.

Before launch I found that every seeded email address used the wrong format, so every notification would have bounced, that the server was running a build two months old, and that local test runs were writing demo data into production. All three were fixed before anyone logged in.

After launch

Every comment goes into a backlog in the person's own words, with who said it and when. I shipped two updates within five days, wrote the notification rules as a table plus a 9-case test script, and turned email on for one tester before turning it on for everyone.

To help people actually use it, I made a 14-section manual with 39 screenshots, four short walkthrough videos and an in-app What's New page.

How I used Claude

Claude Code co-authored every commit. The project's instructions include a rule I added after Claude once told me the database was backed up when it wasn't: never state anything that hasn't been checked. Mistakes are written down in the project log so they don't happen twice.