HomeTaste

Inspiration

Sharing a home does not mean sharing the same taste. A room update can stall between different preferences, an existing sofa nobody wants to replace, and a limited budget. HomeTaste explores a practical decorating adviser that helps two people find common ground before they buy anything.

What it does

HomeTaste takes two taste profiles, a room, a goal, and a budget. Users can add measurements, furniture to keep, and household needs, or skip the optional interview. Three fictional sample households let judges test the full flow without entering personal information.

The workflow resolves favorites to real Qloo entity IDs, requests separate Home Goods brand recommendations for each person, prioritizes overlap, and explains each candidate's rank. Selected Qloo aesthetic tags inform an illustrated composition of coordinated lighting, textiles, and small accents.

Users can veto a brand or request warmer finishes, lower spending, or easy-care materials. HomeTaste replans while retaining the room, existing items, and budget ceiling. A visible planning trace and downloadable brief explain what came from Qloo and what HomeTaste inferred.

How we built it

The app uses React, TypeScript, Vinext/Vite, and a Cloudflare Worker hosted as a public web app. Qloo /search resolves taste references; /v2/insights supplies Home Goods brands and aesthetic tags. The Qloo key stays in a server-side runtime secret. Names, measurements, budgets, and household notes are not sent to Qloo.

The current planning agent is a bounded deterministic workflow with explicit validation, tool calls, ranking, composition rules, and a feedback/replanning loop. It does not use an LLM or an agent framework. Private server-side caching reduces repeat Qloo requests, and the public demo includes a best-effort per-instance rate limit.

The source is MIT-licensed. The README includes setup and judge testing instructions.

Challenges we ran into

Cultural affinity is not the same as furniture fit. Qloo supplies useful home-brand signals and aesthetic metadata, but not a verified inventory of furniture, prices, or room dimensions. We therefore show illustrative item briefs and budget allowances, not purported product listings or accurate room reconstructions.

We also needed to handle households whose taste lists have no shared brand. HomeTaste explicitly identifies that case instead of presenting a false compatibility score.

Accomplishments that we're proud of

We shipped a working public end-to-end demo with real Qloo integration, four room types, optional dimensions and interview, retained furniture, three fictional test households, coordinated item-type boards, and constraint-preserving revisions.

Type checking, the production build, and eight automated tests pass. Tests cover ranking, vetoes, validation, room-brief retention, fictional inputs, budget accounting, and changes caused by taste tags and revision requests.

What we learned

Qloo makes cross-domain discovery concrete. In a live test, Bob Dylan and West Elm led to Anthropologie appearing in both recommendation lists. Different taste references produce different brand candidates and aesthetic inputs; removing that evidence would remove the household-specific discovery layer.

This is a technical demonstration, not a claim that cultural taste reliably predicts every decorating preference or that we have outperformed an LLM-only baseline. Transparent evidence and small, reversible room decisions are more useful than confident guesses.

What's next for HomeTaste

Next are richer household negotiation, better composed room visualizations, verified item sourcing with real dimensions and prices, optional photo interpretation with explicit uncertainty, and user testing against generic decorating advice. Automatic measurement from a single uncalibrated photo and affiliate links are outside this prototype.

Judge quick start

1. Open the public demo; no sign-in is required.
2. Choose Music meets modern to load a fictional bedroom, an $800 lighting budget, an existing bed/tables, and Bob Dylan / West Elm taste profiles.
3. Click Find our common ground and compare the live Qloo directions.
4. Inspect the coordinated item brief, retained constraints, and planning trace.
5. Try Warmer finishes or Spend about 25% less, then veto a direction and download the plan.
6. Compare with Small home office to test a different room and budget.

Illustrations are conceptual, amounts are planning allowances, and actual prices, availability, delivery, and furniture fit are not verified.

Built With

• qloo