Burrito.ai

My household has someone with a tree nut allergy and someone who is lactose intolerant. I want to share how I use the allergen features to manage this. Every ingredient in my library has allergens tagged. When I view a recipe, the allergen summary shows at the top: 'contains: dairy'. When I am planning a meal for the whole household, I filter recipes by 'no dairy, no tree nuts' and only see what is safe for everyone. The most useful part: the allergen check runs on sub-recipes too. My mole sauce is a sub-recipe in three other dishes and it has a nut content note. Every dish that uses the mole sauce shows the nut allergen flag without me having to remember manually.

I want to write a specific use case for version history because I think people do not know about it. I had a carnitas recipe that had been working perfectly for a year. I made what I thought was a minor modification — changed the fat trim percentage I mentioned in the preparation step. The next batch was noticeably different and I could not figure out why. I went to the recipe history (three-dot menu > Version history), compared the current version to the version from three months ago, and found that in addition to the fat trim note I had also changed a temperature value from 130C to 160C without remembering it. That was the actual cause. I reverted to the 130C version and the results are back. The history is there — use it.

Six weeks in, two trucks live, writing this up because I could not find anyone who had done it. What worked. Roll completion events into the POS as line item confirmations. Recipe IDs mapped to POS product codes with a lookup table we maintain by hand, about two hundred rows. Nightly reconciliation job that compares generated against sold and flags gaps. What hurt. There is no bulk endpoint for recipe metadata, so building the lookup table meant two hundred individual calls and a rate limit I discovered the hard way. Clock skew between the POS and the event timestamps put things on the wrong side of midnight until we normalised everything to UTC on receipt. And the sandbox does not emit roll completion events at all, so the entire event path had to be tested in production against a truck that was not open yet. Would do it again. Would budget three weeks rather than one.