The Recipe Finder Got Picky, and Started Saving the Keepers Itself

By Blane Warrene · · building-in-public, healthy-cooking, recipe-discovery, usability

When I first wrote about the recipe finder, the pitch was simple: tell it what’s in the fridge and it goes hunting through cooking sites I trust, then hands me a ranked shortlist. It worked. The problem, a few weeks of real use later, was that the shortlist was too generous. It showed me everything it found, including recipes that matched one ingredient out of five. This month I taught it to be choosier, to do more of the filing itself, and to read one more site.

A grid of preview images from the eleven cooking sites the recipe finder reads: Serious Eats, Preppy Kitchen, Billy Parisi, Barefoot Contessa, Gordon Ramsay, Jamie Oliver, Bon Appetit, Rainbow Plant Life, Love and Lemons, Brian Lagerstrom, and Food Network.

It hides the near-misses now

Every recipe the finder pulls gets a score, and the score is mostly one thing: how many of my ingredients the recipe actually uses. A lemon-and-chicken search turns up a lemon-herb chicken (high score) alongside a lemonade (matches “lemon,” nothing else). The lemonade used to sit there on the list anyway.

Now there’s a floor. By default, anything that covers less than about a third of my ingredients drops off the shortlist before I ever see it, and the finder tells me what it dropped and why. Nothing vanishes silently, it just moves to a short “didn’t make the cut” note at the bottom. When I want to browse wide instead of narrow, I can lower the floor or switch it off entirely with a single setting.

The effect is small to describe and large to live with. I would rather skim five recipes I would genuinely cook than wade through twenty and do the filtering in my head.

It saves the keepers without me

Saving used to be a tiny back-and-forth every time: the finder lists what it found, I pick the numbers I want, it files them as drafts. Fine for a handful, tedious when the search is good and most of the list is worth keeping.

So I gave it a threshold I can set: anything scoring at or above this bar, just keep it. The strong matches now land in my drafts folder on their own, and I review only the borderline ones by hand. Less clicking, the same amount of control.

The control part matters, so I want to be precise about it. Saving is not publishing. A saved recipe is a private draft, and promoting a draft into my real library is still a separate, deliberate step I take on purpose. Auto-save speeds up the part that was busywork (capturing candidates) without touching the part that should stay slow (deciding what becomes real).

One more cook at the table

The finder only reads a hand-picked list of sites I already trust, and for a while that list had a hole in it. Food Network kept the door shut: it blocks anything that looks like a robot, even though its own posted rules say the recipe pages are open to read. I found a polite way back in that respects those rules, so it is reading again. That brings the list to eleven:

Two things keep the list honest. It leans toward how I actually eat, so seafood, chicken, and vegetables float up while beef and pork drop out automatically. And when one of these cooks writes a recipe as free-flowing prose instead of a tidy machine-readable card, the finder can still read the ingredient list, facts only, always credited back to the source. It never reproduces anyone’s method or passes their words off as mine.

It also stays a good guest on the way through: it honors each site’s rules about what may be read, waits a beat between requests so it is never hammering anyone, and remembers what it already fetched so a repeat search is instant and doesn’t re-bother the site.

Where this is heading

The theme of the last few updates has been the same, even when the features look unrelated: make this thing quieter and more trustworthy, so I reach for it without thinking. A shorter shortlist, a folder that fills itself with only the good stuff, one more trusted source. None of it is flashy. All of it is the difference between a tool I poke at and one I actually cook from.

The shortlist is the whole point, and it is finally what I wanted: short, mine, and ready to cook. The next thing I want is to stop typing at it and start talking to it, the way I already talk to the meal logger. That’s the experiment I’ll write up next.

I keep writing about this as it grows, what works and what surprises me. If that’s your thing, subscribe. Roughly weekly, no filler.