Olivv Tester: Internal AI Ops Console

The problem

Olivv ships primarily through a mobile app, which is great for customers and slow for iteration. Testing whether a new image model produces better campaign posters, or whether the in-app assistant handles a booking flow, meant clicking through the app, watching server logs, and keeping cost estimates in a spreadsheet. Comparing models was mostly guesswork.

Who it’s for

The Olivv team: engineers testing generation endpoints, product folks reviewing what the scheduler produced before it reaches customers, and anyone deciding which image model is worth its price.

How it works

It puts the entire backend surface in one browser tab: image and campaign generation, content review queues, restaurant onboarding, a posting-strategy planner, and a live dashboard. Every generation shows what it cost. A batch run sends the same prompts across every account, model, and generation type and returns a pass/fail grid, which turns model comparison into a table instead of an argument.

The most valuable tab is the least flashy one. The agentic chat loads the same conversation flows that ship inside the Android app and wires them to the real APIs, so the assistant can be tested and improved in the browser before a release goes out.

Decisions worth naming

• Auth mirrors production. If sign-in breaks in the app, it breaks here first.
• Cost is a first-class feature. Editable per-model rates and running totals, because “which model is best” is meaningless without “at what price”.