My Son Needed an Apartment. I Built an AI Platform. Here's What Broke Along the Way.
About this session
This is the story of building Snugd — an AI-powered apartment hunting platform — from personal frustration to live product in six months. It starts with a UPitt grad, a PNC rotation program, and a terrible Google Sheet. It ends with a production app using Claude for 6 distinct AI features, Whisper for voice transcription, Next.js 16 on the frontend, FastAPI on the backend, and a data pipeline scraping 15 cities every 12-48 hours. In between: the prompt engineering that makes Claude return reliable match scores instead of hallucinated enthusiasm, the Whisper integration that turns rambling tour commentary into structured pros and cons, the true cost calculator that exposed how much listings hide, and the multi-channel contact system that works when landlords don't have email. I'll share what worked, what broke, and what I'd do differently if I built it again.
Speaker
Key takeaways
- End-to-end architecture of a multi-AI consumer product: from Apify scraping to Claude scoring to Whisper transcription to Stripe billing
- Prompt engineering patterns for getting reliable, structured AI output in production (match scoring, comparison analysis, decision briefs)
- What breaks when you ship AI to real users — and the fallback patterns that kept the product running