Website Re-Architecture & AI Concierge

A site that was too heavy to measure. Avital Homes' website ran on WordPress with a third-party real estate plugin bolted on for inventory. It looked like a modern builder's site, but underneath it was collapsing under its own weight. The inventory page, the single most important page for a home buyer, delivered nearly 34 MB to every visitor. The homepage delivered 16 MB. For context, a well-built page of this kind should come in under 1.5 MB.

The cause was not a mystery. It was images.

• Every photo shipped at full resolution. Property and gallery photos were served exactly as they were uploaded from the camera, with no compression and no sizing for the screen they were displayed on. One search page alone pulled down more than 31 MB of images across 60 photos. A phone displaying a 400-pixel thumbnail was downloading a multi-megapixel original to do it.
• The entire framework loaded on every page. Whether a visitor was looking at one listing or the whole catalog, the page pulled in 50 to 65 separate scripts and up to 2.8 MB of stylesheets, most of them unused.
• Bloated markup. The page builder generated more than 6,400 elements on the homepage and inventory pages, roughly four times the level Google flags as excessive. On the search page the browser's accessibility tree failed to build at all.
• Mobile was effectively broken. Under Google's standard mobile test, the homepage took 53.7 seconds to show its main content and 57 seconds to become interactive, scoring 34 out of 100. The inventory, community, and listing pages could not be scored at all. The testing tool gave up before they finished loading.

Neither the site nor its data was ready for AI. The inventory lived inside a plugin never designed for a home builder, so critical facts were stored in repurposed fields. The community name lived in a field labeled "Status." The lot number lived in a field called "Year Remodeled." Some homes listed one bedroom on 2,800 square feet. Communities existed under several duplicate web addresses at once. There was no structured, trustworthy source of truth an AI assistant could safely answer from, and the heavy, script-driven pages meant machines could not reliably read the site's content in the first place.

Search visibility was leaking. The site carried hundreds of live URLs, many of them auto-generated archive and taxonomy pages with duplicate content, and no redirect strategy existed. Structured data was missing on the pages that mattered most. Page speed, which Google weighs directly, was working against every ranking the company had earned.

Rebuild, don't patch. Optimizing the WordPress site in place would have meant fighting the plugin, the page builder, and the theme on every page. Instead, Rade Engineering re-architected the site from the ground up on a modern, distributed platform: a React-based framework rendered on the server and delivered from a global edge network, a headless content system the Avital team manages directly, a managed Postgres database with row-level security for customer data, and dedicated object storage for media. Every layer is independently scalable, observable with error and performance monitoring, and secured with defense-in-depth access control rather than plugin permissions.

Make bad images structurally impossible. Rather than asking editors to remember to resize photos, we built an ingest pipeline that does it for them, every time, with no exceptions:

• Every upload is automatically downscaled, converted to a modern web format, and stripped of excess metadata before it is ever stored. Full-resolution originals are never saved and never served.
• The pipeline generates a set of responsive sizes for each image, so a phone receives a phone-sized file and a desktop receives a desktop-sized file.
• Uploads over a sane size limit are rejected with a clear message in the admin, so an oversized file cannot slip through.
• Every image is delivered through a single shared component that requires dimensions and sizing rules, which eliminates layout shift and oversized delivery by design.
• Descriptive alt text is generated automatically by AI at upload when an editor leaves it blank, improving accessibility and giving search engines and AI assistants text to read.

A self-cleaning media library. Real estate inventory churns constantly. Homes sell, communities build out, and on the old site their photos simply piled up forever. The new site runs a periodic scan that identifies every image and document no longer referenced anywhere on the site, quarantines it for 30 days with a "Pending purge" view and a one-click restore in the admin, and then permanently removes it. Shared assets are never touched, because the scan checks what the whole site references, not just what a single listing used. Storage stays lean without anyone having to think about it.

Protect every ounce of SEO equity. A migration is where rankings go to die, so we treated it as its own workstream:

• We built a complete inventory of every legacy URL by combining the old sitemap, 16 months of Google Search Console data, and the site's own database, which surfaced 413 distinct addresses, including ugly permalinks and orphaned pages a sitemap alone would have missed.
• Every one of those 413 URLs was mapped to exactly one new destination, or deliberately retired, with an automated test that fails the build if any address is left unaccounted for.
• Redirects are real, permanent, single-hop responses at the edge. An automated verification run confirmed zero redirect chains and zero loops, and that the top 50 pages by search impressions all resolve correctly.
• Duplicate community and floorplan pages were consolidated onto one canonical address each, removing the duplicate-content signals that had been diluting rankings.
• Every page now emits structured data for search engines (listings, floorplans, communities, business information, breadcrumbs, FAQs), per-page metadata is editable in the CMS, and the site publishes a clean sitemap and an llms.txt file so AI search tools can understand the business as clearly as Google does.
• Analytics continuity was preserved by reconnecting the existing Google Analytics and Tag Manager properties, so reporting history was unbroken across the cutover.

A site that is dramatically faster on every page, measured with the same tools against the same pages.

• Inventory page weight: 33.9 MB down to 1.4 MB
• Homepage weight: 16.0 MB down to 1.2 MB
• Community page weight: 9.8 MB down to 0.8 MB
• Homepage elements: 6,465 down to 458
• Desktop performance score: 74 up to 98
• Mobile performance score: 34 up to 72 to 85
• Desktop time to main content: 2.9 seconds down to 1.1 seconds

Pages are 13 to 24 times lighter across the board. Desktop performance scores sit between 96 and 98 on every page type, with zero layout shift and zero blocking time. The three pages that could not even be scored on the old site because the test tool gave up now score in the mid-70s to mid-80s on a simulated slow mobile connection. Content pages are pre-rendered and cached at the edge, so the server responds in a fraction of the time it used to.

Search equity carried over intact. Every legacy URL resolves to its correct new home with a single permanent redirect. Nothing 404s. Structured data, editable metadata, a live sitemap, and clean canonical URLs give the site a stronger SEO foundation than it has ever had.

Inventory that maintains itself. When a home closes, its page redirects to its community automatically, its heavy media is retired, and a "homes we've built" card takes its place. When a community sells out, it stays live as a lean, indexed page rather than vanishing. Unused images clean themselves up on a 30-day clock.

And now there is an AI concierge. Because the inventory, communities, floorplans, and promotions live in one clean, structured system, the site can now offer something the old one never could: an always-on assistant, available on every page, that answers buyers' questions from live data. It knows which homes are actually for sale right now, what they cost, what floorplans and elevations are offered, which communities have which amenities, and what promotions are running. It only answers from real inventory, never from guesswork, and it is aware of the page a visitor is on, so "tell me about this one" just works. When a visitor is interested, the concierge captures their details and routes them to the sales team in seconds. The same clean data that makes the site fast is what makes the AI trustworthy.