AI concierge on live listings
Answers after-hours buyers on a coastal platform’s own inventory — ranked matches, a written reply, a human loop. The proof is above.
Fixed-scope AI builds — production systems, live in 30 days
The buyer who messages at 9pm shouldn’t have to wait until morning.
We build production AI systems: fixed scope, fixed price, live in 30 days. Flagship in production — an AI concierge that answers after-hours buyers, matches homes, and lines up the showing.
Concierge live in production on real listings
Fixed scope · Fixed priceLive in 30 daysThe spec is yours to keepA named acceptance test
Proof, not a demo
Built and measured on a live platform.
We do not demo on toy data. The concierge below runs on a client’s real inventory, against real buyer traffic.
Client anonymized by agreement: a California coastal real-estate platform.
See it work
2:14am. The phone rings. Nobody’s awake — except the concierge.
The narrated cut, 63 seconds with sound: the voicemail arrives, the concierge answers on the platform’s own listings, matches homes, and lines up the showing.
Fictional brokerage and listings — the behavior is the product. Client surfaces anonymized by agreement.
The exchange, from the cut
2:14 am · voicemail
“Hi, it’s late — we’ve been looking online all week, and there’s a bluff house we keep coming back to. I think it had a big west view? We want to know if it’s still available, and what it would take to see it. Call me tomorrow, no rush.”
overnight · concierge
“I found 3 demo listings for ‘Show me homes where I wake up to water’. I ranked them by fit, not just by matching words.”
→ 12 Bluffton Crest · $4.85M · 4 bd · 4.5 ba — leads the set on the water view
by morning · concierge, to the agent
“Thanks for calling Aureline Estates — we got your message overnight. … The strongest match is 12 Bluffton Crest in Windward Bluff at $4.85M (4 bd · 4.5 ba). There are 3 matches worth a look — I can send the full shortlist. Shall I have Jordan Vance call you and line up a showing?”
Excerpt from the narrated cut — fictional brokerage and listings; the behavior is the product. The build’s eval rail graded the exchange pass · confidence 0.92.
Live
AI concierge in production
Real buyers, real listings — not a staged demo or sandbox.
96 / 96 / 100 / 100
Lighthouse, live production
Performance, accessibility, best practices, SEO. Median-of-3.
40.5MB → 6.8MB
Hero media ladder
A 76-byte inline placeholder holds first paint; weight climbs only when it must.
SSIM ≥ 0.97
Quality enforced in CI
Video quality is a tested invariant. Re-encodes must re-pass or the build fails.
- Progressive media, by designImage rungs climb 960w to 1920w; video loads when the browser is idle; data-saver devices get the poster and stop.
- Zero third-party requestsSelf-hosted variable fonts, immutable caching at the edge — no font CDN on the critical path.
- Guarded by testsA unit and structural battery plus end-to-end browser tests and Lighthouse floors. The build fails before a regression ships.
Selected work
The flagship above — and the rest of the shelf.
Different domains, same discipline: production systems, fixed scope, measured against acceptance tests.
Medlyze — price-transparency copilot
Plain-English questions into BigQuery answers and visuals, with guardrails tuned to regulated data.
Svod — resume search that cites itself
Natural-language candidate search with source attribution and instant results, built for HR teams.
Voiux — transcription, self-hosted
A production voice pipeline behind a simple REST interface, speaker diarization included.
Znahar — D2C wired into social
An e-commerce platform built straight into TikTok Shop for a herbal-extracts brand.
How it works
Three steps. One bounded outcome.
No open-ended discovery, no hourly meter. Each step ends in something you can hold.
Map the data and the journey
Where your data comes from, what is stale or missing, and where a high-intent visitor drops off. The gap becomes a number, not a hunch.
Build the assistant on your data
A production concierge on your own listings — answering, matching, and scheduling in your brand’s voice. The workflow goes live, not into a slide deck.
Monitor and optimize
We watch real conversations, fix what falters, and tune matching against the leads that actually convert.
The math
What’s slipping past after hours?
Three honest numbers. The assumptions are written down, and they are conservative on purpose — we’d rather disagree with the figure together than overstate it.
$720estimated commissions slipping past after-hours, per month
Conservative by construction: 25% of after-hours inquiries assumed lost today · 40% of those recovered by the concierge · 3% conversation-to-close.
Pressure-test this on your numbers
The ladder
Fixed scope. Fixed price. Nothing unfunded.
Start with the cheapest yes, or go straight to the build. Every rung names its deliverable and its acceptance test up front.
Paid diagnostic
$1.5–2.5k fixed, one-time
One week. We map your data and the workflow it runs on, size the gap, and hand you a build spec you keep — whether or not you build with us.
Payment: 100% at sign · credited against the build if you go ahead within 30 days
Core build
$7–15k fixed, single workflow
A production assistant on your own data in 30 days. The workflow goes live, with a named acceptance test that closes the engagement.
Payment: 25–30% deposit, milestone-funded
Care plan
$2–4k per month
Monitoring, fixes, and optimization. Converts a one-time build into a system that keeps getting better.
Payment: monthly, cancel any time
Scope is fixed, so price is fixed. Every “one more thing” is a priced change order. We never start an unfunded phase. The diagnostic fee is credited against the build.
Questions
The ones worth answering up front.
What does “fixed scope” actually mean?
One workflow, named in writing, with an acceptance test. If it isn’t in the scope, it’s a priced change order. If it is, we ship it.
What do you need from us to start?
A sample of your data, one decision-maker, and a week. No migrations, no replatforming — the build lives beside what you already run.
Why not an off-the-shelf chatbot?
Generic tools answer generically, and buyers can tell. We build on your own data, measured against your own acceptance test — the flagship does exactly this on a live platform.
Do you only do real estate?
No. Real estate is the flagship, not the boundary — copilots, voice, retrieval, internal tools. If it can be fixed-scoped and tested, we build it.
What happens when something breaks?
It degrades honestly: if a provider is down, the surface says so instead of guessing. Fixes and monitoring come with the care plan; the acceptance test defines “working”.
Who owns what?
The spec from the diagnostic is yours to keep. The build runs on your own data, in your own accounts — the scope spells out the rest in writing.
How fast, really?
Discovery inside a week; a single workflow live in 30 days. The acceptance test closes it — not a slide deck.
What does “Write us” start?
A written exchange. You describe the need; we reply with an honest read of whether it’s worth building — and we say no if the math isn’t there. No calls unless you ask for one.
Tell us your after-hours volume; we’ll tell you if it’s worth building.
Write us a few lines — your after-hours volume, where your listings live. We’ll read it, and say no if the math isn’t there.
Write us
ivan@tnvs.dev