Work

Fresh Supplies (FreshRoute AI)

FastAPI
Next.js
Flutter
PostgreSQL
XGBoost
Docker

AI-assisted post-harvest logistics platform for Sub-Saharan African agricultural cooperatives, combining spoilage forecasting, route recommendation, and multi-tenant operational tooling.

Fresh Supplies Dashboard

AI-assisted post-harvest logistics platform for Sub-Saharan African agricultural cooperatives, combining ML-based spoilage forecasting, dynamic market/route recommendation, and multi-tenant operational tooling to reduce produce lost in transit between farm and market.

Problem

Smallholder farmers and cooperatives routinely lose a large share of harvested produce before it reaches a buyer — not from poor yields, but from logistical blindness in transit: non-refrigerated transport, unpredictable road delays, and zero real-time visibility into whether cargo is still sellable by the time it arrives. Dispatchers have no way to know a shipment is degrading until it’s already a write-off at the destination market.

What It Does

Fresh Supplies tracks produce from harvest through delivery and uses a trained ML model to predict spoilage risk for a shipment based on transit time, distance, and crop-specific perishability. When risk is high, it recommends alternative wholesale markets ranked by revenue retained (quantity × price × (1 − spoilage%)), so a shipment can be diverted to a nearer, better-margin buyer instead of continuing to a market where it’ll arrive unsellable.

Multi-Tenant Role Model

The platform is organized around agricultural cooperatives as tenants, with role-scoped access enforced server-side:

RoleScope
AdministratorPlatform-wide: provisions cooperatives, users, and access grants
Farmer Cooperative (Admin / Member)Owns produce lots and shipments for their own cooperative
Logistics ManagerManages shipments and driver assignment for cooperatives they’ve been explicitly granted access to
Market AnalystViews analytics/market intelligence for granted cooperatives only
DriverMobile-only; sees and acts on shipments assigned to them specifically

Staff roles (Logistics Manager, Market Analyst) get cooperative access via explicit access grants rather than a single fixed assignment, so one person can support multiple cooperatives without being a full admin.

Core Features

  • Produce lot tracking — crop type, grade, quantity, harvest date, per cooperative.
  • Shipment lifecycle — scheduled → in-transit → delivered, with driver assignment and cooperative-scoped visibility.
  • Spoilage prediction — per-shipment risk score from a trained classifier (temperature, transit time, distance, crop frailty).
  • Market/route recommendation — ranks candidate destination markets by distance, time, predicted spoilage, and resulting revenue retained.
  • Mobile driver app (Flutter) — OTP-based login, offline-capable sync of field records, in-app pickup confirmation that flips a shipment to in-transit.
  • Admin dashboard — user provisioning with forced password-change-on-first-login (via emailed OTP), cooperative/tenant management, access-grant management.
  • Analytics & reporting — cooperative-scoped dashboards for shipment and spoilage trends, CSV/report export.

Architecture

Web (Next.js) ─┐
Mobile (Flutter)─┼─▶ FastAPI backend ─▶ PostgreSQL
                 │   (REST + ML inference)
                 │
           Caddy (reverse proxy, path-routed, single origin)
                 │
           Cloudflare Tunnel (named, fixed hostname)
                 │
           Cloudflare edge (TLS termination)
  • Web and backend are served from a single origin behind Caddy (/api/* → backend, everything else → frontend), eliminating CORS in production.
  • The only internet-facing component is an outbound-only Cloudflare Tunnel — no ports are opened on the host; TLS terminates at Cloudflare’s edge.
  • Docker Compose orchestrates Postgres, backend, frontend, Caddy, Cloudflare Tunnel, and one-shot DB provisioning/migration containers.

Tech Stack

LayerTechnology
BackendFastAPI, SQLAlchemy 2.0 (async), Alembic, PostgreSQL 16
FrontendNext.js, React, TypeScript, Tailwind CSS, shadcn/ui, TanStack Query
MobileFlutter/Dart
MLscikit-learn / XGBoost spoilage classifier, graph-based route scoring
InfraDocker Compose, Caddy, named Cloudflare Tunnel
AuthJWT access/refresh tokens, email OTP for first login and passwordless recovery

Status

Actively developed capstone project; core multi-tenant auth, provisioning, shipment/produce management, ML spoilage and market recommendation, and the mobile driver app are implemented and deployed.

Where To Find The Resource