Know if it's still safe to eat — before you take the risk.
SafiFood predicts food spoilage, explains why in plain language, recommends what to do about it, and keeps a verifiable community record — built for households with no sensors, no dedicated app, and unreliable connectivity.
From "is this still good?" to a verified answer
Four steps, the same whether you're tapping through the app or texting on WhatsApp.
Log what you have
Add a food item, how it’s stored, and for how long — in the PWA or over WhatsApp. No sensors, no barcode.
Get a prediction you can trust
An XGBoost model estimates days-to-spoilage, and SHAP explains the top reasons in plain language — five factors or fewer.
Act on it
Approaching spoilage? Get a value-addition recipe — ferment, dry, pickle — matched to that food and that risk level.
Share what you can’t use
List real surplus for your community to claim. Every offer and claim is hashed to Hedera — verifiable, never falsifiable.
Built for the six things that actually matter
Prediction, explanation, value addition, redistribution, verifiability, accessibility — together, not spread across six different apps.
Spoilage prediction
A single XGBoost regressor trained on WHO/FAO guideline data predicts days-to-spoilage from storage conditions — no special hardware required.
Explainable, not a black box
Every prediction ships with a SHAP-backed explanation in plain language, capped at five contributing factors.
Value addition, matched to risk
Approaching spoilage doesn’t mean the bin. Get a preservation technique sized to tools an informal-settlement household actually owns.
Community redistribution
Can’t use it in time? Offer it to your community instead of throwing it away — tracked from offer to claim.
Verifiable, not just trusted
Freshness checkpoints and redistribution events are hashed to Hedera Consensus Service — public, tamper-evident, no PII on-chain.
Meets you on WhatsApp
Alerts and quick replies over the WhatsApp Cloud API, with the full dashboard and history in the PWA when you want it.
Held to specific numbers, not vague promises
Evaluation targets set going into the build — the same ones this project is measured against.
- ≥95%
- Recall target for the spoilage model — missing a real risk is the failure mode that matters
- ≤5
- Contributing factors surfaced per explanation, in plain language
- ≥68
- System Usability Scale target across a 50+ user pilot
- 2
- Hedera Consensus Service topics — freshness and redistribution, hash-only
Stop guessing by smell. Start knowing.
Create a free account and log your first food item in under a minute.
Get started free