Preheat: From Preorders to What to Bake, Buy, and Earn

For home food businesses: turn weekly preorders into what to make, what to buy, and real profit.

• GitHub repo
• GitHub showcase
• Video on Google Drive
• Slides

Problem: Tart Theory, a Seattle home bakery, takes weekly preorders through its website, WeChat and Instagram. Orders live in one Google Sheet and recipes and costs in another, so every week the owner matches payments, counts flavors, scales recipes, builds the grocery list, and guesses at margins by hand. Justification: This isn't unique to one bakery: in JPMorganChase Institute data, only 15.3% of solo businesses with no employees pay for AI tools vs. 26.1% of businesses with employees, and Brookings puts AI adoption in accommodation and food services at just 2.9%. Small businesses already sell across a median of 3 channels (SBE Council), so orders scatter before AI ever helps. What we built: Preheat pulls orders into one place (a webhook that accepts the site's existing order format, CSV import, or a pasted DM that Claude turns into an order) and turns paid orders into the week's work: what to bake per pickup day in whole batches with recipes scaled to grams, a shopping list, a production timeline, a pickup checklist, and Insights on weekly sales, cost and real profit after discounts and Venmo fees. All math is plain code; Claude handles messy messages and drafts bilingual replies, and the baker confirms everything. What makes Preheat different: it was built with a real bakery on real data, with the owner sharing her actual sheets, testing every version and designing the UI herself. Instead of asking bakers to move to a new storefront (with per-order fees) or re-enter orders into a separate costing app, onboarding starts with where they already sell: today a Google Sheet link shows the sheet in Preheat and leads to CSV import, and the bakery's existing website order format is accepted as-is, with one-click connectors for Squarespace, Square, Instagram and WeChat next. Preheat then connects orders to recipes, so paid orders become what to bake, what to buy, and real profit after Venmo fees. Claude handles messy messages; every number is plain code, and the baker confirms every AI result. Built with the owner, who shared her real sheets, tested each version, designed the UI, and named Bake and Insights the most useful parts. Setup: no code. Follow the README to deploy free on Supabase and Vercel in about 30 minutes; the Claude key is optional. Still rough: validated with one bakery; Squarespace, Instagram and WeChat connectors are designed but not built; payments are marked paid by hand; Insights' channel breakdown uses sample data.