DevAI: AI-Powered Deal Finder for Real Estate Investors

Real estate investors in Miami spend hours scanning listings, manually calculating renovation costs, and estimating whether a deal is worth pursuing. Most of the time, the math is back-of-the-napkin at best. DevAI came to Kreante with a clear problem: build a platform that does this automatically, at scale, with AI.

The problem

Single-family home investors need to evaluate dozens of properties before finding one that pencils out. The core calculation — listing price, renovation cost, after-repair value, profit margin — requires pulling data from multiple sources and running estimates that vary by property condition, location, and market comparables.

Doing this manually takes 30-45 minutes per property. A good investor needs to evaluate 50+ properties to find one deal. The math doesn’t work without automation.

What was built

Kreante delivered a functional Proof of Concept in 4 weeks. The system connects to the Zillow API to pull live listings in Miami, runs an AI-powered analysis on each property, and surfaces the results in a clean interface that lets investors filter by profit margin, price, days on market, and square footage.

The core profit formula for the PoC:

Gross Profit = Zestimate (market value) - (Listing Price + AI Renovation Estimate)

OpenAI GPT-4 estimates renovation costs dynamically based on property parameters. The result is stored in PostgreSQL and displayed in a WeWeb interface with a map view and analysis trigger.

Technical stack

PoC (delivered in 4 weeks):

• Backend: FastAPI (Python) with 2 endpoints — GET /properties and POST /analyze/{id}
• Database: PostgreSQL for property data and analysis results
• AI: OpenAI GPT-4 for renovation cost estimation
• Data source: Zillow API (listings + Zestimate)
• Frontend: WeWeb for rapid prototyping
• Infrastructure: Docker + Kamal on Hetzner

Recommended MVP architecture (next phase):

• Backend: Supabase (serverless PostgreSQL + managed APIs)
• Frontend: React SPA built with Lovable
• Data: Bridge Data Output API + multiple MLS integrations
• AI: OpenAI with response caching and batch processing for cost optimization

The switch from a self-hosted Hetzner server to Supabase for the MVP eliminates server management overhead and reduces operational complexity significantly.

What was validated

The PoC hit 100% of its objectives:

• Pulled 20+ live properties from Zillow API with listing price, Zestimate, square footage, GPS coordinates, days on market, and room count
• AI renovation cost estimation working end-to-end (property params in, cost estimate out)
• Profit margin calculated and stored per property
• WeWeb interface operational: map view, property list, “Run Analysis” button, results display
• Lovable prototypes designed for both the investor portal and the superadmin panel
• Dockerized and deployed on Hetzner with Kamal

Two separate UI directions were prototyped and validated with the client during week 4.

The profitability engine

The MVP will extend the PoC analysis model significantly. Instead of a simple Zestimate-based gross profit, the full model calculates:

Net Profit = ARV - (Purchase Price + Renovation Costs + Soft Costs + Financing Costs)

Where renovation costs come from a weighted average of three sources: the AI estimate, standard cost ranges per square foot, and the client’s own historical data. Comparable sales (comps) are pulled automatically — the 5 most relevant sold properties within 1 mile — and the user can remove comps that don’t fit their criteria, triggering an instant recalculation.

Operational costs at scale

Infrastructure costs for the MVP are estimated at $650-950/month:

Service	Cost
Lovable (frontend + hosting)	$25/month
Supabase (database + APIs)	$25/month
OpenAI API	~$200/month
Bridge Data API (MLS feeds)	$400-700/month

The biggest cost driver is MLS data. Optimizing refresh frequency (daily vs. every 3 days) and implementing response caching for the AI layer are the primary levers for keeping costs in range.

What this pattern applies to

The core architecture — API data ingestion, AI analysis engine, profitability calculation, filterable frontend — transfers directly to other deal-flow contexts: commercial real estate, distressed debt, auction platforms, or any domain where humans need to evaluate large numbers of opportunities quickly.

The PoC-to-MVP model used here (4-week PoC to validate assumptions, then a 10-week MVP with refined architecture) is particularly well-suited for investors who want to validate a product idea before committing to full development. Kreante applies this same approach across industries — starting with a contained PoC that de-risks the larger build.