Dining halls in the U.S. waste an estimated $1.8B of food each year. BinSight turns that gut feeling into proof: take one bin photo, and managers see what was wasted, what it cost, and what to change next week.
Over a semester, this can save thousands of dollars and keep hundreds of meals out of the trash.
- Instant waste analysis: Upload a bin photo to get item-level waste, estimated weight, cost, and avoidable vs. unavoidable waste.
- Automatic menu scraping: Pulls the day's menu from the dining hall site so the model knows what was served.
- Cost and waste tracking: Dashboard shows trends over time, category breakdowns, and avoidable/unavoidable split.
- AI recommendations: Uses scan history to suggest concrete procurement and portioning changes.
Create binsight/.env for local secrets and runtime options.
ANTHROPIC_API_KEY=your_key_here
ANTHROPIC_VISION_MODEL=claude-sonnet-4-6
ANTHROPIC_TEXT_MODEL=claude-sonnet-4-6
OPENROUTER_VISION_MODEL=openai/gpt-4o
OPENROUTER_TEXT_MODEL=openai/gpt-4o-mini
RESET_ANALYSES_ON_LOAD=falseRESET_ANALYSES_ON_LOAD=false keeps saved analysis JSON and uploaded images across reloads. Set it to true for demo mode where each page load starts clean. The in-app Reset button still clears saved analyses manually.
LLM_PROVIDER=anthropic uses the Anthropic model settings. LLM_PROVIDER=openrouter uses the OpenRouter model settings. The older VISION_MODEL and TEXT_MODEL env names still work as OpenRouter fallbacks.
Food-waste analysis demo app.