Inspiration
I've been looking at renovation properties in London and the process is painfully slow You find a listing, book a surveyor, wait a week for the report, then get a quote from a builder and you still can't visualise what the finished property could look like. I wanted to compress that entire workflow into a single URL paste. When I saw Gemini 3's multimodal capabilities, especially the combination of Flash Preview's vision analysis and Nano Banana Pro's image-to-image transformation, I realised I could build an AI surveyor that doesn't just identify problems, but shows you the solution.
What it does
PropertyPulse takes a Rightmove.co.uk property listing URL and runs a full renovation analysis in under 2 minutes:
- Scrapes the listing for photos, floorplan, and property details
- Analyses every room photo using Gemini 3 Flash Preview identifying room type, scoring condition (1-10), and spotting specific issues
- Calculates renovation costs using a 2024 UK pricing database with regional multipliers (ie. London gets +35%)
- Generates before/after visualisations using Nano Banana Pro, preserving the original camera angle and room layout
- Refines any visualisation on demand, the user can adjust the prompt and regenerate a new image to their taste.
The entire pipeline streams real-time reasoning steps to the UI, so you can watch the AI think.
How I built it
The backend is a single FastAPI application in Python. There's no frontend build step, just Jinja2 templates with Tailwind CSS and Alpine.js served directly.
The architecture is a four-stage pipeline:
- Scraper (Playwright) navigates to Rightmove, handles cookie banners, extracts images across multiple URL formats
- Analyser (Gemini 3 Flash Preview) - multimodal vision analysis of each room photo
- Cost Engine - applies UK renovation pricing with 12 regional multipliers and categorises work as essential/recommended/optional
- Visualiser (Nano Banana Pro) - style-aware image transformation that detects architectural period (Victorian, Georgian, 1930s, Modern) and adapts the renovation accordingly
Results stream to the browser via Server-Sent Events, giving users a live reasoning feed as the analysis progresses.
Challenges I faced
Scraping Rightmove reliably was harder than expected. They use 3 different image URL patterns across listings, and the cookie consent banner blocks automation. I had to build pattern detection with fallbacks.
Style-aware visualisations took iteration. Early prompts produced generic modern interiors regardless of the property. The breakthrough was detecting architectural style from the analysis and feeding period specific renovation language into Nano Banana Pro so a Victorian terrace gets sash windows and cornicing, not floor-to-ceiling glass.
Keeping the demo under 3 minutes while the full pipeline takes ~2 to 3 minutes meant optimising every delay in the pipeline without hitting API rate limits.
What I learned
The combination of vision analysis and image generation in a single pipeline is genuinely powerful Gemini 3 Flash Preview's ability to assess room condition from a photo, not just identify objects, but evaluate quality and age, was surprisingly accurate. And Nano Banana Pro's image-to-image transformation preserves spatial context in a way that text-to-image never could.
The biggest takeaway: the real value isn't any single AI call, it's the orchestration. Each stage's output shapes the next stage's decisions, making this truly agentic rather than just a wrapper around a prompt.
Sample results
What's next for PropertyPulse?
Multi-platform support — expand beyond Rightmove to Zoopla, OnTheMarket, and international listing sites
Gemini 3 Pro upgrade — leverage the full 1M token context window to analyse entire property portfolios in a single pass, comparing renovation ROI across multiple listings
Video walkthrough analysis — use Gemini 3's spatial-temporal video understanding to analyse property tour videos instead of static photos, catching issues a single frame might miss
Contractor matching — connect cost estimates directly to local tradespeople with real-time availability and pricing
Investment scoring — combine renovation costs with local sale prices to calculate projected ROI and flip profitability
Log in or sign up for Devpost to join the conversation.