Applied AI / Web
MealSpark
Generate recipes from a pantry photo.
About the project
Built during HackUMBC and developed further afterward, MealSpark turns pantry photos and dietary preferences into recipes. The application connects a browser image workflow to a server-side Gemini endpoint.
Implementation decisions
Compress images before sending them to the API to control payload size.
Keep the model key on the server and request a structured JSON recipe.
Separate generation from presentation so the result can be displayed and copied.
Current scope
Recipes are model-generated. The current endpoint forwards requests; production hardening would include stronger input validation, rate limits, and response-schema validation.