AI that respects the numbers.
PropertyIQ uses AI for prose, not projections. The financial numbers are computed deterministically from real market data. The narrative explains them. Here's how the engine stays honest.
Guardrails
Validated against source data
Invented numbers
Reproducible
The discipline behind every report.
Most AI underwriting tools invent comp data, invent occupancy rates, invent ADR. We don't. Here are the six guardrails that keep us honest.
Live data, never cached forever
AirROI, Airbtics, PriceLabs, Apify, Zillow, Rentcast, Furnished Finder. Every data point on every report is timestamped. The Airbnb comp links get a live-status check at publish so if a listing goes dark you'll see 'Archived (data captured April 2026)' instead of a broken link.
Deterministic financial projections
The Conservative / Realistic / Aggressive numbers are NOT written by AI. They're computed deterministically from market stats using locked formulas. Same property + same market data = same numbers every time. Reproducible from input.
AI for narrative only
Claude (Anthropic API) writes the executive summary, the 'why this property outperforms' section, and the 4 tax-strategy plays. Style is locked to a Sotheby's + Bloomberg voice. Output is JSON-schema validated before it touches the DB.
No-invented-numbers guardrail
The AI system prompt explicitly forbids inventing dollar figures, percentages, or comp data. Every number in the narrative must mirror a number already computed in step 2 or 3. If the model tries to invent, the Zod validator drops it and we re-prompt.
Human review pass
Every report sits in 'in_review' status after generation. A Stays Optimized analyst opens it at /portal/admin/engine, reads the prose, checks the comp matches, sanity-checks the projections against their own knowledge of the market, then approves. Only then does it publish.
Owned by Stays Optimized LLC
The methodology, the three-tier framework, the tax-strategy layer, and the report formats are proprietary. Reports are licensed to the customer for their property; they cannot be resold or used to train competing models.
What's under the hood.
| Frontend | Next.js 15 - React 19 - Tailwind 4 - Geist Sans - Framer Motion - Recharts |
| Backend | tRPC v11 - Drizzle ORM - PostgreSQL (Neon) - Better Auth - Inngest |
| AI + Data | Anthropic (Claude) - AirROI - Airbtics - PriceLabs - Apify - Rentcast |
| Delivery | Stripe - Resend - Cloudflare R2 - Vercel |
Order a report. See the engine work.
The fastest way to understand what we built is to run a real property through it. $199 single report, delivered in 10 minutes.