Inspiration
A meal can meet every requirement and still miss the person.
Someone may need an ingredient avoided, prefer a softer preparation, or have only a short time to cook. But the food also needs to feel familiar, appealing and welcome. For a caregiver or family member, remembering those details is part of caring.
That inspired CareTable: a little more care at every meal.
We wanted to bring practical meal planning together with the tastes, traditions and everyday routines that make food personal. Qloo's taste intelligence gave us a way to connect familiar preferences with related dining affinities and make those connections useful in a real planning workflow.
What it does
CareTable helps individuals, partners and families create, understand and improve meal plans.
Users record each person's favourites, dislikes, ingredients to avoid and preparation preference. They choose the planning period, meal slots, available equipment and cooking time. CareTable then brings Qloo taste signals into the selection process and checks the resulting catalog meals against the recorded requirements.
There are two ways to plan:
- User Mode: follow numbered steps, type preferences and choose the planning controls.
- Interactive Mode: talk through the setup with an AI companion, review the proposed details and confirm before a plan is created.
Each meal can explain its fit through a receipt: the resolved taste signals, Qloo's returned dining connections, CareTable's recipe mapping and the applied checks. The receipt also preserves candidate rankings with and without Qloo's contribution.
Care continues after the plan arrives. Chat helps shape a written improvement request; Discuss with AI lets the user explain it aloud. A reviewed, confirmed change revises the selected meal while keeping the other meals and saved requirements in place. Saved plans and their versions make it possible to return to earlier decisions.
A fictional sample provides an immediate way to explore, and email/password accounts support private saved households and plans.
How we built it
We built CareTable with Next.js, React and TypeScript, hosted on Vercel, with Convex managing authentication, saved data and the server-side workflow.
Qloo is the source of the cultural taste connections in that workflow. CareTable resolves suitable food and cuisine preferences into Qloo dining tags, retrieves related affinities through Qloo Insights, and maps those returned connections to its original recipe catalog. The strongest mapped affinity contributes to candidate ranking, so repeated synonymous tags do not multiply a cuisine's influence.
Those signals have a concrete job: helping determine which eligible choices rise to the top. Qloo provenance and the ranking comparison remain attached to saved meals, making the contribution inspectable.
OpenAI supports conversational interpretation, live voice and selection among eligible recipe IDs. CareTable's application code coordinates the process and validates exclusions, equipment, cooking time, ownership and completeness. The CareTable Meal Guide turns the reviewed inputs and Qloo-backed candidates into a plan.
We authored 26 prototype recipes for the project. Convex Auth and Resend support account verification and recovery; server-side environments protect provider credentials. GitHub pushes to the main branch deploy through Vercel.
Challenges we ran into
The central challenge was bringing taste discovery and firm requirements together without confusing their roles. A cultural affinity should help rank an eligible choice; it should never override someone's recorded ingredient exclusion.
We also needed to represent different people at the same table. Requirements remain tied to their intended recipients, with independently checked portions available when a shared meal does not fit everyone.
Voice introduced another important design challenge: conversation must lead to a clear decision. We built editable reviews, explicit confirmation, retained conversation drafts, interruption guidance and deliberate retry options. A dropped connection should not force someone to repeat everything or apply a change they have not reviewed.
Accomplishments that we're proud of
We built a complete hosted journey from preferences to a saved plan, visible Qloo evidence and a focused meal improvement.
In a controlled nine-scenario development comparison, Qloo changed the top-three ordering in six scenarios and changed 12 meal slots in deterministic control plans, while preserving the eligible sets. That gives us a measurable demonstration of Qloo influencing selection.
Our recorded hosted walkthrough also follows a spoken fish request through review and confirmation to a saved 35-minute fish meal within a 45-minute allowance. Read-back checks confirmed that the other meals and original requirements were unchanged.
We are proud of making the experience feel caring as well as understandable: numbered guidance, contextual mode descriptions, warm visual design, colour-coded statistics and a choice of typing or speaking. The finished English-captioned walkthrough runs 2 minutes 19 seconds.
What we learned
Personalisation becomes more useful when people can see why a choice was made and still shape what happens next.
We learned to give each part of the system a clear responsibility: Qloo connects tastes, conversational AI helps people express what they want, and application checks protect the recorded requirements. Bringing those parts together creates a workflow that people can inspect, confirm and improve.
We also learned that recovery is part of the experience. Keeping a draft, explaining the next action and preserving the previous plan are practical ways to show care.
What's next for CareTable
We want to broaden the recipe catalog and reviewed cultural mappings, add reliable regional ingredient and price information, refine accessibility across mobile and voice interaction, and extend usage reporting for larger audiences.
The goal remains the same: make thoughtful meal planning easier to understand, easier to revisit and more personal to the people around the table.
Thank you, Qloo, for the taste intelligence at the heart of CareTable's discovery workflow.
CareTable. A little more care at every meal.
Built With
- convex
- convex-auth
- css3
- ffmpeg
- html5
- javascript
- next.js
- node.js
- openai
- playwright
- python
- qloo
- react
- resend
- typescript
- vercel
- vitest
- zod
Log in or sign up for Devpost to join the conversation.