[CASE // REILU]
Sustainable grocery companion
Snap a photo of a product or receipt. Reilu identifies items, looks them up on Open Food Facts, estimates health and environmental impact, and suggests better more sustainable alternatives. Utilizes a multi-agent AI pipeline to keep failures isolated and missing data flagged rather than hallucinated.
The story
Built for a hackathon focused on using AI for the good of humanity. Microsoft wanted a product utilizing a multi agent approach and our solution was Reilu. Grocery packaging aren't always forth coming about health scores, eco-ratings, carbon footprint and shoppers are left none the wiser.
Reilu bridges the gap between shoppers and the data they need to make informed decisions. Users no longer to be be professional nutritionists or environmental scientists to understand the impact of what theyre buying. Reilu.
How it works
A photo upload triggers four specialized agents in sequence. Each has a narrow job and structured output, so failures stay isolated and missing data is flagged rather than hallucinated.
The Vision Agent extracts items from the photo. The Facts Agent verifies each item against Open Food Facts by barcode, structured search, or full-text fallback. The Summary Agent produces health and eco scores plus a human-readable summary. The Suggestion Agent finds better swaps in the same category for low-scoring items.
An Impact Enricher runs alongside, attaching category-level CO₂e estimates from Climatiq (or a local fallback table). Missing impact data returns unavailable — never an hallucinated value.