README.md

metadata

title: Order Triage
emoji: 🏆
colorFrom: green
colorTo: blue
sdk: gradio
sdk_version: 6.18.0
python_version: '3.13'
app_file: app.py
pinned: false
license: mit
tags:
  - backyard-ai
  - track:backyard

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

📦 Order Triage Assistant

Backyard AI track — Build Small Hackathon

The problem

At Zipline's KD-1 distribution centre (Kaduna, Nigeria),many drone delivery orders for medical supplies and vaccines arrive as informal, abbreviation-heavy text messages via Telegram/social channels — e.g.:

Pcm syrp 4
Pcm drop 3
Pcm tabs 1 pkt
Given set 1
Scalvein 10 mixed 23g and 21g

Marmara Hc
Vaccines requesiton tomorrow 11/06/2026 Thursday please
BCG 20 doses
Opv40
Penta 30
...

Fulfillment staff then manually re-type each line into the inventory/order fulfillment system — slow, error-prone, and easy to misread abbreviations (is "Scalvein 10 mixed 23g and 21g" two items or one?).

What this app does

1. Paste the raw order text as received.
2. A small LLM (Cohere Command, ≤32B class) extracts structured data: sender, facility, requested date, urgency, and a list of line items (item name, quantity, unit), expanding common abbreviations.
3. Each item is fuzzy-matched against your product catalogue (with SKU codes), so "PCM syrp" → "Paracetamol Syrup 120mg/5ml" with its real SKU.
4. Low-confidence matches are flagged for human review rather than guessed silently.
5. Output is shown as an editable table and exported as JSON (ready to feed to an inventory API) and CSV.

Catalogue

This Space ships with a small sample catalogue (sample_catalogue.csv) covering the product categories seen in real KD-1 orders (analgesics, vaccines, syringes, cannulas, diluents, etc.) — built for demo purposes without exposing any internal Zipline data.

To use with a real catalogue, upload your own CSV with columns: sku_code, product_name, category, unit

Setup

1. Get a Cohere API key (free tier credits work fine).
2. Set it as an environment variable COHERE_API_KEY, or paste it into the "Settings" panel in the app (not stored).
3. pip install -r requirements.txt
4. python app/app.py

Demo

Here is the link to the demo

Social Media Post

Here is the X post on my submission

Roadmap / next stretch

• Voice input for low-literacy users.
• Direct push to inventory system API (instead of JSON export).