CrumbDesk: AI Can Read the Cake Order. It Doesn’t Get to Run the Bakery

Description and use

CrumbDesk uses a local Gemma 3 4B model via Ollama to extract structured data from informal customer messages, giving bakers a reviewable workspace to confirm orders and manage changes with a detailed history.

Problem

Home bakers need a way to turn informal customer messages into structured, reviewable order details while preserving baker authority over accepted orders.

Outcome

CrumbDesk enables bakers to reliably extract, review, and manage orders with AI while keeping control over final decisions and preserving a clear change history.

Workflow

• 1. Baker copies a customer message into CrumbDesk.
• 2. CrumbDesk sends the text to local Gemma 3 4B via Ollama for structured extraction.
• 3. The extracted data is parsed, normalized, and displayed in an editable review panel.
• 4. Baker reviews, edits missing details, and confirms the order or drafts a clarification reply.
• 5. Confirmed orders are stored; change requests become proposals that require explicit baker approval.
• 6. All actions are logged locally and can be exported as CSV.

Evidence

CrumbDesk turns manually copied customer messages into structured, reviewable order details using Gemma 3 4B through local Ollama inference.

Source

The app does not automatically read WhatsApp or Instagram messages.

Source

CrumbDesk validates the distinction between quantity and weight, preserves unresolved date wording rather than guessing, and prevents payment claims from becoming payment verification.

Source

Every important action is recorded locally: creation, confirmation, change requests, decisions, and field differences.

Source