Used Car Finder
A daily used-car search for one buyer in the family. It scores every car within 250 miles of Panama City against his criteria and shows the math.
A family member needed a car and the rest of us were hunting for him. This does the searching every morning. A paced pull from Auto.dev covers dealers. A scraper running in real Chrome on my MacBook reads Autotrader, Craigslist, Facebook Marketplace, and OfferUp overnight, and anyone in the family can paste a link they found. It’s tracking about 2,600 cars right now.
Every car starts at 50 and gains or loses points on his criteria: an estimated out-the-door price against a $23,000 target, the model, year, miles, color, cloth seats, history, and how hard it’s been driven. The car page shows each line of that math. Claude Haiku reads the descriptions, photos, and dealer pages for the top cars to fill in equipment, and every feature is marked confirmed, probably, no, or unknown, with where that came from. Listing data is wrong a lot (placeholder prices, stock photos, the wrong color), so the app labels what it doesn’t trust instead of hiding it. Price gets checked against mileage-adjusted comps, the KBB range from Autotrader, and a Claude web search, and a car gets a “Good deal” badge when two of them agree.
The family can flag cars and comment, and he votes each one up or down. I’ve revised the criteria twice since launch. He’s down to a couple of cars now, and a deal sheet puts the out-the-door math, his loan quotes, and the comps for each on one page to take to the dealer.
Every car starts at 50. Each line of the score says where its points came from.