KitchenPal: Your AI Kitchen Companion Inspiration It started with a simple question that's surprisingly hard to answer: "What should I eat?" Research shows we make 227 food decisions every day, and couples spend over 5 days per year just deciding what to cook. This "decision fatigue" doesn't just waste time—it drains our willpower, leads to unhealthy choices, and turns cooking from a joy into a chore. We asked ourselves: What if AI could eliminate this friction entirely? What if you could just open your fridge, snap a photo, and have a friendly chef guide you through making something delicious with exactly what you have? That's KitchenPal. What It Does KitchenPal is an AI-powered kitchen companion that transforms how people approach cooking:

Fridge Vision — Snap a photo of your fridge, and Google Gemini's vision AI identifies all your ingredients instantly Smart Recipes — Get personalized recipe suggestions based on what you actually have, not what a recipe assumes you should buy Voice Coach — Tap any cooking step, and our ElevenLabs-powered voice coach explains it like a patient, friendly chef standing beside you—not a robot reading instructions Beautiful Imagery — Every recipe features AI-generated food photography via Gemini Imagen, making dishes irresistible before you even start

How We Built It Tech Stack LayerTechnologyFrontendNext.js 14, React, TypeScript, Tailwind CSSAI Text & VisionGoogle Gemini 1.5 FlashAI Image GenerationGoogle Imagen 3Voice CoachingElevenLabs Text-to-SpeechDatabaseSupabase (PostgreSQL)StorageSupabase StorageDeploymentVercel Architecture Highlights Google Gemini Integration:

gemini-1.5-flash for vision analysis (fridge scanning) and recipe generation imagen-3.0-generate-002 for photorealistic food photography Structured JSON responses for reliable recipe parsing

ElevenLabs Voice Coaching:

Custom coaching prompts that sound encouraging, not robotic Context-aware guidance: "Alright chef, let's get started!" for step 1, celebratory tone for the final step Real-time audio streaming for instant feedback

Smart Caching:

Generated images persisted to Supabase Storage for fast loading Recipe caching to minimize API calls Rate limiting to handle usage spikes gracefully

Challenges We Faced

  1. Making Voice Feel Human Our first voice implementation read instructions like a GPS—accurate but lifeless. We iterated extensively on prompt engineering to make the coach sound like a supportive friend: ❌ "Step 3: Add salt to taste" ✅ "Nice work! Now let's season it—add a pinch of salt, taste, and adjust until it's just right"
  2. Image Generation Quality Initial Gemini image outputs were inconsistent. We solved this with detailed "professional food photography" prompts specifying lighting, plating, angles, and color grading—resulting in magazine-quality visuals.
  3. Fridge Vision Accuracy Getting Gemini to reliably identify ingredients required careful prompt engineering. We built a two-mode system: dish mode for identifying prepared foods, and ingredients mode optimized for raw ingredient detection in refrigerators.
  4. Handling the "Recipe Not Found" Flash On page reload, recipes briefly showed "not found" before data loaded. We traced this to React hydration timing and fixed it by properly chaining loading states between the data hook and UI. What We Learned

Voice UX is underrated — The difference between a robotic reader and a supportive coach is entirely in the prompt design AI image generation needs guidance — Generic prompts produce generic results; detailed creative direction unlocks quality Decision fatigue is real — User testing confirmed that removing the "what to cook" decision genuinely reduced stress Multimodal AI is ready — Combining vision, text, image generation, and voice creates experiences that felt impossible just a year ago

What's Next for KitchenPal

Dietary preferences — Respect allergies, restrictions, and nutritional goals Pantry memory — Track what you have so you don't need to snap every time Meal planning — Weekly suggestions that minimize waste and maximize variety Shopping integration — One-tap ordering for missing ingredients Hands-free mode — Full voice control for when your hands are covered in flour

KitchenPal: Because "what's for dinner" shouldn't be the hardest question of your day.

Built With

Share this project:

Updates