Add hackathon README

README.md

@@ -12,4 +12,63 @@ license: mit
 short_description: Turn messy DMs into clean orders.
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# DM Order Desk
+
+DM Order Desk helps tiny sellers turn messy customer messages into a clean order sheet, prep list, and reply drafts.
+
+It is designed for home bakers, farmers market vendors, food truck operators, and small Instagram or WhatsApp sellers who take orders through direct messages instead of a full ecommerce system.
+
+## What It Does
+
+Paste messy customer DMs into the app. The app extracts:
+
+- customer name
+- item
+- quantity
+- flavor or variant
+- pickup time
+- pickup place or delivery address
+- payment status
+- missing details the seller still needs to ask for
+
+It then generates:
+
+- a structured order sheet
+- a prep list for fulfillment
+- short customer reply drafts
+
+## Example Use Case
+
+A home baker receives several messages:
+
+```text
+Maya: Hi! Can I get 2 dozen cupcakes for Saturday morning? Half vanilla, half chocolate.
+Sam: Need 1 birthday cake, chocolate, for pickup Friday 5pm. I can pay Venmo.
+Lena: Do you still have lemon bars? I need some for tomorrow but not sure how many yet.
+Chris: 12 cookies please, pickup at the farmers market. Paid already.
+```
+
+DM Order Desk turns these messages into a structured order table, a prep list, and follow-up replies for missing details.
+
+
+## Why Small Models Fit
+
+This is a narrow, practical workflow. The model does not need broad world knowledge or long-form reasoning. It only needs to extract structured order details from short messages.
+
+The app uses:
+
+- Model: `Qwen/Qwen2.5-1.5B-Instruct`
+- Parameter count: about 1.5B
+- Total model size: well under the 32B hackathon limit
+- UI: Gradio
+- Hosting: Hugging Face Spaces
+
+## Track
+
+Backyard AI
+
+This project is built for a real everyday problem: tiny sellers often receive orders through messy DMs and need to manually turn them into something they can fulfill.
+
+## Limitations
+
+This is a prototype. It may still need human review for ambiguous messages, unusual products, or complex multi-message conversations. The goal is to reduce manual sorting work, not replace seller judgment.