One receipt photo into Notion: turn each purchase into stock and unit price, then plan meals from what is actually there
D I G IX (Twitter) · 2026-09-21
Read originalConnected @muse with my @NotionHQ meal planner. Now I can take a picture of my grocery receipt have each purchase registered to an ingredient in my pantry, Updating my pantry stock, tracking per item cost, and suggests meals based on what I actually have in stock vs my recipes. I also had it update all the broken/empty image links in my Recipes database so my gallery view is fully populated now 🤩
Business model
The author connected Muse to his Notion meal planner. He photographs a grocery receipt and every purchase is registered to an ingredient, updating pantry stock and tracking per-item cost, with meal suggestions based on what is actually in stock versus his recipes. He also had it repair the broken and empty image links in the recipes database so the gallery view filled in. No figure appears in the post and none is invented here. What transfers is the chain that turns physical stuff into data: receipt, line items, ingredient entries, stock and unit price, then meals. Household waste and duplicate buying both come down to not knowing what you already have, and once stock and price are structured, what to cook first and what is quietly getting more expensive become arithmetic instead of guesswork.
The leverage is not buying less, it is knowing what you already hold. Once stock and unit price are structured, what to eat first and what to stop rebuying stop being impressions and become calculations. The side task is worth copying too: repairing broken image links is exactly the kind of maintenance nobody wants to do and everyone notices when it is done, which makes it ideal to hand over whole.
Workflow
- Build the ingredient library first, with canonical names, units and aliases, before wiring any receipt into it.
- Have the agent parse the receipt into fixed fields and mark anything it cannot read as unidentified rather than guessing a likely product.
- Map each line to an existing entry and update stock, with unclear conversions flagged instead of silently applied.
- Record the unit price every time and keep the last few, because a single price tells you nothing.
- Compare the updated stock against the recipes and return three tiers, now possible, missing one or two things, and needs a dedicated shop, with items near expiry first.
- Have it list the bad links, empty images, missing entries and duplicates as a repair list you approve before anything is changed.
Adaptation plan
The post gives no accuracy rate, no number of entries, no amounts and no outcome. Whether the parsing was reliable enough to trust as a record, and whether it changed how the household shops, is unknown. Recognition quality depends heavily on receipt legibility, and thermal fading, folds and glare all degrade it quickly.
Starter prompt
AgentFlow starter draft: not tested in the original business setting; results are not guaranteed. Synthetic examples do not verify the author’s revenue.
You are my pantry inventory assistant. I do not want a menu. I want what I actually have turned into reliable data, and meals planned from that. You do not overwrite or delete anything in my library without my confirmation. [INPUT: what I will provide] 1) One or more photographs of a grocery receipt. 2) My ingredient library structure: fields, units, and the canonical names and alias table I already use. 3) My recipe library structure, including the ingredients and quantities each recipe needs. 4) A snapshot of current stock, if I have one. 5) My dietary constraints: allergies, dislikes, number of people and a rough per-meal budget. [STEPS] 1. Parse the receipt into fixed fields: item name, quantity, unit price and line total. Anything you cannot read is UNIDENTIFIED. Do not guess a likely product and do not merge things that merely look like the same category. 2. Match each line to an existing ingredient using my alias table. Anything unmatched goes to NEW_ENTRY_PENDING with a suggested name and unit, and waits for me. Do not fold it into a similar entry on your own. 3. Update stock with conversions applied: ingredient, this change, resulting quantity, unit, and the basis for the conversion. Ambiguous conversions stay PENDING_MY_CONFIRMATION. 4. Record the unit price for each item and keep the last few observations, marking the direction of change versus the previous one. With fewer than three observations you make no statement about prices rising or falling. 5. Compare updated stock against the recipes and return three tiers: possible now, missing one or two things, and needs a dedicated shop, with anything near expiry first. Every suggestion must state which stock it consumes and how much. 6. Sweep the library for bad data: dead links, empty images, ingredients with no entry and duplicate entries. Give me a repair list and I decide what changes. [OUTPUT] - Table 1: receipt parsing, with matched entry and unidentified items. - Table 2: stock updates, with quantity, unit and conversion basis. - Table 3: cost record, with this price, previous price, direction and observation count. - Table 4: meal suggestions, with tier, stock consumed, gaps and whether anything is near expiry. - Closing: the repair list plus an UNIDENTIFIED and PENDING list. [CONFIRMATION REQUIRED BEFORE CONTINUING] - Unidentified items and pending entries enter the library only after I confirm. - I perform every write, overwrite and deletion. You produce the change list only. - Prices and quantities follow the receipt. You do not estimate, round or substitute typical prices. - A price trend may be stated only after three or more observations of the same ingredient, and you cite my own records as the basis. - Allergies and dislikes follow my input. You do not relax them or decide what I can eat.
Running costs and pitfalls
The post gives no accuracy rate, no number of entries, no amounts and no outcome. Whether the parsing was reliable enough to trust as a record, and whether it changed how the household shops, is unknown. Recognition quality depends heavily on receipt legibility, and thermal fading, folds and glare all degrade it quickly. No independent cost benchmark is available. Check subscriptions, permissions and external actions before running a workflow.
