How to build an expense policy checker
An expense policy checker that reads each claim against your own policy, explains what it found, and sends the ambiguous cases to a review queue. Six prompts, and the last one deploys it to Tray Helix.
Why it matters
Most expense review is a person reading a receipt next to a policy PDF to confirm a claim that was fine all along. A policy checker does the reading. Clear claims go straight through, and the rest arrive at a reviewer with the rule they touched and the gap already written out.
Your policy is specific to you, and it changes. Keeping it in one editable file, with a checker built around it, is now a short build with an AI coding assistant. Tray Helix runs the result behind single sign-on, with a named owner, which matters because expense data is personal data.
What you’ll see
- The finished expense policy checker on Tray Helix, with reviewers working from their own queue
- The policy the checker reads from, on its own tab
- The six prompts from the Helix guide: the policy as an editable file, a check that explains each finding, a review queue and exception report, a pre-submission check for claimants, and connecting your own systems
- The last prompt, “deploy my app”, returning a live URL
- The deployment record showing who shipped it
Chapters
Related
Transcript
Read the transcript
This is an expense policy checker running on Tray Helix. Reviewers work from a queue of their own, and the policy it checks against has its own tab.
We built it with an AI coding assistant, from six prompts in a guide anyone can copy.
Vibe-coded apps die on laptops. This one runs on Tray Helix, the governed runtime for AI-built apps. One command put it in production. It sits behind single sign-on, and it has a named owner. Here's how to build it.
Start with the guide. It gives you six prompts in order, and each one leaves you with something that runs.
Prompt one puts the policy in a file you can edit, with nothing hard-coded. Two reads each claim and explains what it found, instead of a pass or fail. Three builds a queue for the ambiguous claims and a report on every exception. Four lets a claimant check a draft claim before submitting it. Five shows how to connect your own systems.
The sixth prompt is three words. Your assistant runs one command, and the app comes back as a live URL. Every deployment is on record with who shipped it.
A SaaS expense tool usually works the vendor's way. An app you build checks claims against your own rules, explains every finding, and changes with a prompt.
Get the six prompts from the guide on the Helix site. Your people build it, Helix runs it.