Case study · concept work, clearly labeled
A tower that doesn't exist, selling the way real ones should.
Solmara Residences is a fictional 92-home development on a real, empty point of Virginia Key, Miami. We built its entire presale platform — end to end — to show developers what launch day can feel like.
The problem developers actually have
A new development doesn't get its construction loan for renderings — it gets it for presale velocity. Yet most launch sites are brochures: beautiful, inert, and blind. They can't show inventory moving, can't answer a buyer at 11pm in Spanish, and give the sales team nothing to run in the gallery. The lender sees a PDF, not demand.
What we built instead
- The revealA scroll-driven arrival: dawn mist, a blueprint ghost, 41 floors surfacing from the Atlantic, ~1,000 windows igniting at night. Fully procedural — no renders, no video.
- The unit explorerAn orbitable model with all 92 residences live — filters dim the tower in sync, every home has a fact sheet, a schematic plan, and a true view from its floor.
- Launch dayA demoable opening: inventory flips reserved and sold in real time with buyer-city activity — watch it move. This is the moment that convinces a lender.
- Sol, the conciergeFive languages, voiced, and grounded in the live inventory. Ask in English, follow up en español — or in العربية, עברית or ייִדיש — mid-conversation; her chips select homes in 3D and open the hold flow.
- The reserve flowRefundable hold + private-tour scheduling, patterned on the best-converting tour-request flows, ending in a reservation agreement sent for e-signature.
- The private sideBuyer and broker portals: payment milestones, a construction journal, documents, shareable residence links, client registration.
- The gallery kioskThe explorer installs as a fullscreen PWA with an offline shell — the iPad in the physical sales center runs the same platform.
What's real and what isn't
Every page carries the concept label. This is the line-by-line version, because a studio that blurs it on its own work will blur it on yours.
| Element | Status | Detail |
|---|---|---|
| The brand, tower, prices, buyers | Fiction | Solmara does not exist. No unit is for sale, no price is an offer. |
| The site | Real | An empty shoreline on Virginia Key, Miami (25.7422, −80.1442), verified against public map data. |
| The skyline across the bay | Generated impression | Downtown Miami and Key Biscayne sit at their true bearings (307.9° and 197.4°) and true angular size, but the buildings are AI-generated art, not photographs of the actual skyline. Distances are compressed ~0.36× so the bay fits inside the render's depth range. |
| Sky, clouds, facade material | Generated source art | Cloud forms and the curtain-wall material were generated, then rebuilt into engine textures. Everything else in the scene is procedural geometry. |
| The constellations | Real shapes, placed for the shot | Orion, the Plough, Cassiopeia, Scorpius and Canis Major are drawn from their real geometry and real magnitudes — Betelgeuse and Antares are red because they are red, Sirius is the brightest because it is. Where they sit on the dome is a composition decision, not an ephemeris: this is not the sky over Miami on any particular night. |
| The rooms behind the glass | Simulated, not modelled | There are no interiors. Each of the 1,066 panes ray-marches the view direction into a virtual box one pane deep and shades it by which face it hits — a ceiling, a back wall, a blind. Building 1,066 real rooms would be pure overdraw for something a few pixels across at reveal distance. |
| The code, data engine, five-language craft | Real | ~4,000 lines of first-party code, reproducible from this repository. |
| The performance numbers below | Measured | Lighthouse runs, re-measured on every visual release. Reports included. |
| The projection further down | Assumption chain | Labelled as such, with every input shown. Not a forecast. |
Measured, not claimed
| Page | Performance | Accessibility | Best practices | SEO | Vitals |
|---|---|---|---|---|---|
| The reveal | 85 | 100 | 100 | 100 | LCP 1.4s · CLS 0 · TBT ~450ms |
| Unit explorer | 100 | 100 | 100 | 100 | LCP 1.4s · CLS 0.01 · TBT 60ms |
Lighthouse 13.4.1, headless Chrome, measured against this live URL
on 2026-08-04 after the five-language pass. Both figures moved, in opposite
directions, and both are published as measured. The explorer went 95 → 100; the reveal
went 92 → 85, the median of eight runs spanning 81–89.
Two things are tangled in that reveal number and it would be easy to report only the
flattering half. The honest split: about three points are ours, established by running
the current build and the pre-i18n build alternately on one machine and reading the paired
deltas (−3, −2, −2, −5, −1) rather than comparing two batches, which on this hardware
invents differences that are not there. The rest is not like-for-like — the older figure was
Lighthouse 12, and this is 13, which scores differently. LCP actually improved,
1.8s → 1.4s; the reveal is held down by TBT, which is three.js compile cost and not
something the language work touched. That the explorer carries the same enlarged dictionary
and a larger concierge and still scores 100 is the evidence for that reading — so the fix is
not to shrink the dictionaries, and we have not pretended otherwise.
Treat these as a band, not a decimal: the run-to-run spread on this hardware is wider than the
number of digits suggests, and two of fifteen runs during this pass came back catastrophically
low from transient machine load and were discarded as outliers, not averaged in.
Accessibility is 100 on every page again — during this pass it was found to have quietly been
96 on the reveal and 95 on the portal since the studio credit shipped in August, because that
credit was dimmed below the contrast floor, and 98 on the explorer, which had never had a
<main> landmark. Both fixed. Payload: 344KB of Arabic and Hebrew type was
added, and an English visitor downloads none of it — the faces load only for the language that
uses them. Reports in docs/lighthouse/.
~4,000 lines of first-party code. Two libraries (three.js, Leaflet — vendored). No framework and no build step. The reveal carries 624KB of generated imagery (clouds, palms, curtain-wall material, distant skyline); every form in the scene — tower, island, water, city — is procedural geometry lit in real time. 60fps on the animation loop.
If this were your launch
Projection — an assumption chain, not a measurement:
A funded launch campaign bringing 5,000 qualified visits in week one · industry interactive-demo engagement (~50%) → ~2,500 people inside the model · 2% requesting holds → ~100 hold requests with tour times attached · at an 8–12% hold-to-contract rate → 8–12 contracts in the first weeks — the velocity evidence a construction lender wants to see. Your numbers will differ; the platform exists to find out fast.
Process
Six milestones, each demoed before the next began: the reveal → the explorer → launch day + holds → the concierge → the portals + kiosk → this evidence pass. Design decisions were grounded in pattern research across real products (tour-request flows, rate-table ledgers), and every milestone shipped with verification, not vibes.
Want this for a development that exists?
Solmara is the concept. The platform — reveal, explorer, launch simulation, multilingual concierge, portals, kiosk — is the product, and it re-skins to your tower, your inventory, your brand.
© 2026 Rivlux Media. Solmara Residences is a self-initiated concept project — a fictional development on a real map point; any resemblance to real developments is coincidental. Built and measured July 2026.