NutrifyAI
Point a camera at food, get a nutrition breakdown. NutrifyAI uses Gemini’s multimodal inference to identify dishes from a single photo and return calories and nutritional insight — no barcode, no manual search.
01 What problem does it solve?
Manual food logging is the reason diet tracking fails: searching databases and estimating portions takes longer than eating. A photo is the lowest-friction input a user can give.
02 How does it work?
food photogemini multimodal inferencestructured nutrition output
- Image intake: Pillow handles upload processing before inference.
- Multimodal prompting: a structured prompt instructs Gemini Vision to identify the dish and estimate nutritional content in a consistent format.
- Instant response: results render in the Streamlit UI in one round trip — the entire product is a single vision-language call engineered well.
- Config hygiene: dotenv keeps API credentials out of the codebase.
03 What did it achieve?
1 photoinput needed
visionLLM modality
livestreamlit cloud
A deliberately small system that shows multimodal LLM engineering: one well-structured vision call replaces an entire manual data-entry workflow.
04 Stack
Multimodal LLMGemini Vision APIPythonStreamlitPillow