Kind Kitchen

Inspiration

Parents for Community Living’s challenge inspired us to make everyday cooking easier to follow for adults with developmental disabilities. A recipe can seem simple while assuming someone already understands measurements, kitchen tools, and words like “shred” or “sauté.”

We wanted to make that hidden knowledge visible through respectful guidance that supports choice, confidence, and learning.

What it does

Kind Kitchen guides users through choosing a recipe, gathering ingredients and tools, reviewing skills and safety notes, and following one small cooking action at a time.

Users can search for recipes, save guides on their phone, hear instructions aloud, and use optional voice commands such as “next step” or “repeat.” Timers sound when time is up. Tappable cooking words open explanations, while visual lessons cover techniques and measurements, including teaspoons, tablespoons, and millilitres.

Allergy labels and trusted-helper checkpoints support preparation for higher-risk tasks.

How we built it

We built Kind Kitchen with React Native, Expo, and TypeScript, sharing app code across mobile previews and a public browser demo.

Gemini creates structured recipe drafts and transcribes short voice commands. ElevenLabs provides natural narration, with device speech as a fallback. Our hosted API keeps provider keys off the phone and checks recipe structure before displaying a guide.

Tiger Data stores shared recipe copies, photo attribution, and aggregate cache statistics through a hosted PostgreSQL bridge. Cloudflare D1 provides the first cache lookup and installation-specific search history, while R2 stores searched dish photographs.

Cloudinary optimizes reference photos, with bundled images available if delivery fails. Reusing completed recipe requests reduces repeated AI calls while preserving differences in toppings, dietary requests, and serving amounts.

Challenges we ran into

Our biggest challenge was shortening instructions without removing essential measurements or safety details. We also needed to explain unfamiliar words without losing the user’s cooking step or timer.

Recipe search introduced spelling mistakes, slow responses, usage limits, and similar dish names. Shared caching and conservative spelling corrections helped address these problems.

We also worked through image-loading issues and layouts where captions obscured photos. Coordinating narration, microphone input, and network transcription required careful handling; voice control remains optional, with buttons available throughout.

Accomplishments we’re proud of

We connected recipe discovery, visual learning, spoken guidance, timers, and local saving in one working prototype.

The iPhone preview and natural narration have been tested on a real phone. We verified live Tiger Data recipe reuse and Cloudinary delivery, alongside 266 automated tests.

What we learned

Accessibility involves the whole interaction: wording, pacing, pictures, navigation, and recovery when something goes wrong. Short instructions still need precise measurements and clear responsibilities.

We also learned that automated checks cannot determine whether a recipe is appropriate for a particular person. Human review and testing with intended users remain essential.

What’s next

We want to review Kind Kitchen with Parents for Community Living and intended users, improve recipes and visuals through their feedback, and expand physical-device accessibility testing.

We also plan to improve voice reliability and prepare standalone iOS and Android distribution. Kind Kitchen remains a hackathon prototype that needs partner and intended-user testing before independent real-world cooking.