Ruvaan · Furniture and appliance rental
Ruvaan rents furniture and appliances in six cities. Its support team answers the same questions all day from a pile of documents: the catalogue, city pages, policies, memos that change those policies, and a lot of everyday notes. Your job is to build them an assistant that answers from those documents.
Download Ruvaan's documents (the zip below). They are yours alone: someone else's figures won't match them. Assume today is 15 September 2026. A dated memo overrides the policy it changes, and a proposal applies to nobody.
Build it however you like, in any language, with any model, AI tools included. Put it online at a public https address (a free host is fine) that answers a POST with a JSON body {"question": "..."} with JSON {"answer": "...", "sources": ["policies/deposit.md"]} within 30 seconds.
Keep each answer to a sentence or two (under 300 characters). In sources, list the paths in the zip your answer came from. When the documents don't answer a question, reply with "answer": null and an empty sources list. Don't guess.
When you hand in, DayOne asks your assistant 20 questions you haven't seen, one at a time, and marks its replies. Keep it running until you've handed in and for an hour after. Also hand in your repository and a walkthrough video of 3 minutes or less: how it works, and the choices you made.
The data is yours alone: someone else’s answers won’t fit it. Use any tool you like, AI included.