- Problem
- 10 to 20% of adopted pets go back to shelters, often because of a poor match.
- My role
- Solo builder. Product, design, and code, with AI tools.
- What I built
- An adoption app that recommends pets based on a person's preferences and lifestyle.
- Results
- Shipped an MVP fast with a tight scope. Retired in December 2025.
The problem
Too many adoptions don't work out. Between 10 and 20% of adopted pets go back to shelters. Mismatched expectations and behavior issues are common reasons.
A University of Alberta study found 80% of adopters choose mostly on looks rather than fit. Meanwhile, 34% of pet owners buy from breeders, and 47% wrongly believe shelters only have older or poorly trained animals.
I adopted my dog Shasta and saw how hard the process is. I wanted fewer returns by making better matches from the start.
What success looked like
The north star was successful adoptions: pets who find permanent homes. On day one, clicks on the Adopt button were the signal I could measure.
Key decisions
A focused MVP shipped fast beats a perfect product shipped late. So I made hard cuts:
- Three animals, not dozens. The PetFinder API covers dogs to chinchillas. I launched with dogs, cats, and rabbits.
- Filters upfront. Filtering happens in the API call, not on the results page. Simpler first version.
- Photos first. Pets without photos rank lower, because photos drive the emotional connection.
- Two paths. A guided questionnaire for people who want a recommendation. Quick Search for people who just want to browse.
- One AI pick. Results show a grid with a single highlighted recommendation, to keep the choice easy.
How I built it
Next.js 14 on Vercel, the PetFinder API for live adoptable pets, and OpenAI (GPT-4o mini) for recommendations. PetFinder needed OAuth 2.0 through a Cloudflare Worker proxy. Getting structured, parseable recommendations took careful prompting.
I built it alone with v0, Cursor, Windsurf, and Claude. Getting to 70% was surprisingly easy. The last 20 to 30% (polish, edge cases, advanced features) took about 80% of the time. My best tip: have the AI document its work as it goes. You see what went wrong, and you can point it to the docs instead of re-explaining everything.
Results
The MVP shipped with a tight scope and tracked Adopt clicks from launch. Next on the list were richer lifestyle questions, a top three with comparisons, funnel analytics, and follow-ups to confirm real adoptions.
I retired FindMePet in December 2025, after the PetFinder API it depended on went away.
Takeaways
- Tight scope ships. Three animal types and one AI pick got a working product out fast.
- Set the north star before launch. Adoptions were the goal. Adopt clicks were the signal I could measure on day one.
- PMs can build now. With AI tools, I went from idea to working app in a fraction of the usual time.