Sufra

Inspiration

For a single man who didn't learn to cook until his late 20s, the idea of mastering the kitchen felt daunting, especially while navigating life in Germany far from Syrian roots. Sufra was born from this personal struggle, leveraging my AI skills in Gemini 3 to create a cultural bridge—turning pantry scraps into authentic Arabic feasts without overwhelming beginners.

What it does

Sufra is an AI-powered culinary engine using Gemini 3 Flash with structured JSON (responseSchema) for "constrained culinary analysis." It generates recipes using only user-available ingredients while computing nutritional density and cost-per-serving in real-time.

Gemini’s multimodality powers the UX: Users upload pantry photos for automatic ingredient detection via Gemini Vision. Gemini 2.5 Flash Image then generates culturally accurate, professional dish photography, making suggestions visually irresistible.

Beyond recipes, Gemini Grounding with Google Search verifies techniques and links to authentic video tutorials. "Market Run" mode uses Google Maps to dynamically locate nearby specialty grocers (halal butchers, spice souqs) via geolocation—turning gaps into actionable shopping. [

Key Features:

• Multi-Input: Photo analysis, text entry, Arabic autocomplete, 30+ staple quick-adds
• Recipe Modes: Pantry-only, Market Trip (2-3 extras), 7 cuisines (Syrian, Lebanese, etc.), budget caps
• Personalization: Persistent "My Shelf," seasonal awareness, allergen detection
• Grounded UX: Maps for suppliers, verified tutorials, bilingual RTL English/Arabic
• Visuals: AI-generated food photography with shimmer loading

How we built it

Built solo in Google AI Studio as a no-code prototype, using Gemini 3 Flash.

Challenges we ran into

The need to go back in history when a prompt didn't work as expected.

Accomplishments that we're proud of

Achieved real-time nutritional/cost analysis from photos alone, preserving authentic Syrian flavors in 100% pantry-constrained recipes like mujaddara variations. Seamless integration of vision-to-visuals-to-grounded maps makes it demo-ready for hackathons and real kitchen use.

What we learned

Gemini 3 excels at agentic workflows like sequential vision-to-recipe-to-maps, but precise system prompts are key for cultural accuracy in niche cuisines. User testing highlighted mobile-first needs; offline caching for staples boosts accessibility in low-connectivity areas like aid scenarios. Structured JSON prevents hallucinations better than free-text generation.

What's next for Sufra

Develop a full mobile app with user accounts for "My Shelf" persistence, difficulty sliders adjusting steps (e.g., beginner voice guides), and seasonal/budget personalization. Integrate voice input for hands-free cooking and expand to crisis modes with aid-kit presets, tying into Syria health tech goals..

Built With

• gemini
• googleaistudio