I built a camera-first property intelligence app solo, nights and weekends. It's live. Here's everything so far.

Point your phone at any house in America → owner, how long they've held it, equity position, distress signals, and a 1–10 deal score in seconds. There's a Walk Mode: hold the phone up and every building you face identifies itself as you move. That's MRA Scout.

Stack: vanilla JS single-page app (no framework), Netlify serverless functions, Supabase for auth/sync/RLS, Claude Haiku reading the user's photo for property condition, licensed property-data APIs behind a swappable one-file vendor adapter.

The build lesson that mattered most: the hard part of a "camera app" is not the camera. It's what your product does at the edge of what it knows. My first data vendor returned the wrong street's owner flagged as a successful match — an app trusting that flag shows a stranger's finances for the house you're standing in front of. I audited eight counties, documented every failure, switched primary vendors through the adapter in one file, and built a rule I now consider the product: an unknown renders as "unknown," never as a guess. My early scoring bug treated a missing mortgage as 100% equity — "we don't know" displayed as the strongest buy signal in the model. Never again.

Honest numbers: launched this month, $0 MRR (Stripe finishing this week), 14K organic impressions on the launch post, ~100 profile views, three data-company execs replied to cold outreach including my vendor's CEO (strategy call booked). Pricing: $49/mo, first 500 lock it forever. Costs ~$250/mo to run; ~10¢ of data per scan against $49 subscriptions.

What I'm figuring out now: distribution to real estate pros — a high-trust, low-software-adoption audience that lives in Facebook groups and REIA meetups, not on LinkedIn. Demo converts great; reaching them repeatably is the puzzle.

AMA — especially happy to go deep on property-data vendor quality (I have benchmarks), the iOS compass permission trap in Walk Mode, or pricing a B2B tool for solo operators.