• Home
  • About
  • Voice AI
  • AI Infrastructure
  • Consulting
  • Claude
  • Training
  • Websites
  • Industries

Follow Us

AGXNTSIX
  • About
  • Voice AI
  • Infrastructure
  • Consulting
  • Claude
  • Training
  • Industries
LoginContactContact
Commercial Real Estate

How Arena District Turned 14 Road Signs Into a 24/7 Bilingual AI Sales Funnel

Arena District, a 156-acre commercial real estate development near McAllen, Texas, hired Agxntsix to turn 14 road signs into a 24/7 sales channel. Agxntsix built a bilingual AI phone platform on Retell AI — receptionist Sofia and sales agent Marco — that answers every call in English or Spanish, syncs to Twenty CRM, and warm-transfers hot buyers with full caller-ID privacy.

Agxntsix Team

Updated on Jun 2026

Client
Arena District
Industry
Commercial Real Estate Development
Location
Hidalgo / McAllen, Texas (Rio Grande Valley)
Scale
156 acres, 11 CBD-zoned lots, 14 physical signs
Services
Voice AI, AI Infrastructure, Marketing Site
Timeline
Ten-week build, March–May 2026
Status
Live in production at 956-917-LAND
Arena District AI Infrastructure Case Study
Case Study
  • Overview
  • Challenge
  • Solution
  • Results
  • Deep Dive
  • FAQ

Overview

Arena District is a 156-acre commercial real estate development with 11 CBD-zoned lots in Hidalgo, Texas, in the McAllen-Edinburg-Mission metro of the Rio Grande Valley. Fourteen physical signs display one vanity number — 956-917-LAND — and every call to it lands in a platform Agxntsix engineered: two named Retell AI voice agents (Sofia, the receptionist; Marco, the sales specialist), a TypeScript/Hono webhook server, a 10-step pipeline into Twenty CRM, Claude post-call analysis, a five-tier drip sequence, Twilio SMS, and the Astro marketing site at arenadistricttx.com. The engagement ran from first commit on March 10, 2026 through production hardening into May 2026.

The Challenge

Fourteen signs along Rio Grande Valley roads generate inbound calls from developers, investors, and brokers at all hours — many in Spanish. Before the platform, every inquiry depended on a human being free to pick up, and a missed call on a multi-acre commercial lot is a missed seven-figure conversation with no record it ever happened.

  • Sign-driven inquiries arrived around the clock, but answering depended on a person being available at that moment
  • Spanish-speaking prospects — a major share of the McAllen-area market — had no guaranteed bilingual coverage
  • No systematic record of who called, which of the 11 lots they asked about, or what was discussed
  • Follow-up was manual and ad hoc, so warm leads cooled off without a nurture sequence
  • Connecting a hot buyer to the broker meant giving out a personal cell number with no privacy layer

The before-state was a signage funnel with no infrastructure behind it: a prospect drives past one of the 14 signs, dials, and the outcome depends on luck — who answers, what language they speak, whether anyone writes it down. After-hours and weekend calls simply evaporated, each one an unrecoverable relationship with no trace left behind.

Why Agxntsix

Agxntsix custom-engineered the platform rather than configuring an off-the-shelf answering bot: version-controlled agent prompts, 48 automated test files, TCPA consent tracking with unsubscribe endpoints, A2P 10DLC-registered SMS, and a TREC-compliance page on the public site. Real estate communication is regulated, and this build treats compliance as architecture, not an afterthought.

The Solution

Agxntsix deployed two named voice agents on Retell AI. Sofia, the AI receptionist, answers every inbound call 24/7, auto-detects English or Spanish on the caller's first words, classifies intent across six categories — general info, parcel inquiry, pricing, showing, broker transfer, directions — and runs a structured intake. Marco, the AI sales specialist, takes qualified buyers deeper into lot pricing, backed by a Retell knowledge base of property facts. Hot prospects are warm-transferred to the broker through a caller-ID-masking privacy bridge. After every call, a TypeScript/Hono webhook server runs a 10-step pipeline: parse, normalize phone, extract analysis, write to Twenty CRM, score the lead, log a note, create tasks, send email, send SMS, enroll in drip.

Key Features
  • Sofia: bilingual AI receptionist with automatic English/Spanish detection and six-category intent classification
  • Marco: AI sales specialist with a get_lot_details tool and a Retell knowledge base covering the 11-lot parcel matrix
  • Warm-transfer privacy bridge that masks caller ID so the broker's personal number is never exposed
  • 10-step Twenty CRM pipeline with a deduplicating SQLite job queue and exponential-backoff retries
  • Claude post-call transcript analysis with Zod-validated structured output, emailed to the broker within 60 seconds
  • Five-tier Claude-generated drip email campaign spanning Day 1 to Day 30
  • A2P 10DLC-registered Twilio SMS texting property packets in the caller's language within 60 seconds
  • Astro marketing site at arenadistricttx.com with an interactive parcel map, 12-article resources blog, and Meta Conversions API lead tracking

Implementation

The first commit landed March 10, 2026, and the platform was hardened in production through May 2026 — 235 commits of iterative tuning, including voicemail handling, a pronunciation dictionary, and NATO-phonetic readback for contact details. Agent prompts are version-controlled and deployed by script, 48 Vitest test files cover the system from webhook queue to CRM filter contract, and the server runs behind a Cloudflare Tunnel with graceful degradation.

The most differentiated piece is the warm-transfer privacy bridge. When Sofia or Marco identifies a hot prospect, Retell AI bridges the live call through a dedicated backline number while the prospect sees only the branded Arena District caller ID. The broker joins in real time — introduced with the caller's name, language, and the parcels discussed — and his personal cell number is never revealed. It converts the highest-intent moment of the funnel into a live human conversation without sacrificing the principal's privacy.

Reliability is engineered in, not assumed: webhooks land in a WAL-mode SQLite queue with uniqueness enforced per call and event, so duplicates never create duplicate CRM records; failed jobs retry on exponential backoff; and a token-bucket limiter keeps Twenty CRM traffic under 90 requests per minute.

Technology Stack

TypeScript
Hono
Retell AI
Twenty CRM
Anthropic Claude (Haiku / Sonnet / Opus)
Twilio SMS (A2P 10DLC)
SQLite (WAL mode)
Astro
Cloudflare Tunnel
Zod + Vitest

The Results

Every call generated by 14 physical signs is now answered 24/7 in English or Spanish, logged to Twenty CRM with a structured record of the lots discussed, and followed up automatically — shipped from first commit to production hardening in roughly ten weeks.

MetricBeforeAfter
Inbound call coverageAnswered only when a person was free; after-hours calls evaporatedSofia answers 24/7 on 956-917-LAND
Spanish-speaking callersDependent on a bilingual human being availableAutomatic language detection, full Spanish parity
Lead recordsNo systematic record of callers or lots discussed10-step pipeline writes every call to Twenty CRM with lead score, note, and tasks
Hot-buyer handoffRequired sharing a personal cell numberWarm transfer with caller-ID masking
Follow-upManual and ad hocSMS packet within 60 seconds plus a five-tier drip, Day 1 to Day 30

Additional Outcomes

  • 956-917-LAND vanity number live across 14 physical signs, all routing into the platform
  • 235 commits and 48 automated test files from March 10 to May 18, 2026
  • Claude transcript analysis emailed to the broker within 60 seconds of every hang-up
  • 12-article SEO/AEO resources blog and interactive parcel map live at arenadistricttx.com

These are infrastructure facts, not projections: the call-analytics database holds full transcripts and structured analysis for every production call — which lots, which language, which intent — visibility that did not exist in any form before the platform.

How does a bilingual AI phone agent handle Spanish-speaking callers?

Sofia, Arena District's AI receptionist, detects the caller's language on the first turn and conducts the entire conversation in English or Spanish — every prompt, intake question, and guardrail exists in full Spanish parity. Property packets, SMS messages, and drip emails are then delivered in the language the caller actually spoke.

Bilingual coverage is not a translation layer bolted on afterward: both agents' prompts were written with full Spanish parity as a hard rule and hardened in production so neither drifts back to English mid-call. In the McAllen market, that is the difference between qualifying a lead and losing it on the greeting.

How does an AI receptionist transfer calls without exposing a broker's personal phone number?

The platform uses a warm-transfer privacy bridge: when Sofia or Marco identifies a hot prospect, Retell AI bridges the call through a dedicated backline while the prospect sees only the branded Arena District caller ID. The broker joins the live conversation without his personal cell number ever being exposed.

The transfer is warm, not blind: before the human joins, the AI announces the caller's name, language, parcels discussed, and a summary of the need — captured during Sofia's structured intake, which must complete before any transfer fires. A tuned agent-swap delay makes the hand-off sound natural rather than abrupt.

What happens to call data after an AI voice agent hangs up?

Within seconds of hang-up, a SQLite job queue runs a 10-step pipeline: parse, normalize phone, extract Retell analysis, write to Twenty CRM, score the lead, log a note, create tasks, send email, send SMS, and enroll in drip. Claude analyzes the transcript and emails the broker a summary within 60 seconds.

Core CRM writes must succeed or retry with exponential backoff, while secondary steps are isolated so a transient email failure never blocks the record of the call itself. Post-call analysis extracts clean canonical CRM values, including 11 boolean fields marking exactly which Arena District lots the caller expressed interest in.

What's Next

Nurture-sequence drafts for Twenty CRM workflow automation are already staged in the repository alongside a workflow CLI, and the queue-based architecture — agents, pipeline, drip, and SMS — is templated so additional developments or phone lines can be onboarded onto the same platform.

Frequently Asked Questions

Can an AI phone agent answer calls in both English and Spanish?

Yes. Arena District's Retell AI agents auto-detect language on the caller's first words and switch instantly. Sofia handles reception and intake in either language, Marco discusses lot pricing in either language, and follow-up emails, SMS property packets, and the five-tier drip sequence are all delivered in the language the caller spoke.

How long does it take to build an AI voice system for a real estate development?

Agxntsix took Arena District's platform from first commit to production hardening in roughly ten weeks, March through May 2026. The core webhook server, both voice agents, and CRM pipeline shipped first; warm transfers, SMS, drip automation, and the marketing site followed in weekly iterations — 235 commits and 48 automated test files in total.

What is a warm transfer with caller-ID masking?

A warm transfer means the AI agent stays on the line and introduces the prospect before connecting a human. Arena District adds caller-ID masking: Retell AI bridges calls through a private backline and displays only the branded company number, so the broker takes hot leads live while his personal cell stays private.

Can AI handle text follow-up after a phone call?

Yes. Arena District runs an A2P 10DLC-registered Twilio number that texts callers a property packet within 60 seconds of hanging up, in English or Spanish to match the call. The five-tier drip email sequence then nurtures the lead from Day 1 through Day 30 with Claude-generated content.

Can an AI receptionist recognize returning callers?

Yes. The platform recognizes returning callers before the first ring finishes. A pre-call lookup races a two-second timeout against Twenty CRM, then hands Sofia dynamic variables — name, preferred language, last topic discussed, and parcels mentioned — so repeat prospects are greeted personally instead of restarting intake from zero.

Medical Aesthetics & Beauty

Turn Your Inbound Calls Into a 24/7 Sales Funnel

Agxntsix builds bilingual voice AI platforms that answer every call, log every lead, and hand hot prospects to your team — privacy intact. Book a consultation to see this architecture applied to your business.

Schedule Consultation

Turn Your Inbound Calls Into a 24/7 Sales Funnel

Agxntsix builds bilingual voice AI platforms that answer every call, log every lead, and hand hot prospects to your team — privacy intact. Book a consultation to see this architecture applied to your business.

Schedule Consultation

Spearheading the AI Evolution

Singapore | Dubai | United States

Member of the Anthropic Partnership Program

Follow Us

LinkedInTwitterYouTubeInstagramFacebook

Stay Connected

AI insights delivered weekly

Solutions

Voice AIAI InfrastructureOpenClawConsultingClaude ImplementationAI TrainingWebsite DevelopmentAnswer Engine OptimizationLead Generation Marketing

Industries

Exotic CarsYachtsPrivate JetsService BusinessesReal EstateE-commerceHealthcareFinancial ServicesGovernmentLegal ServicesView All Industries →

Resources

BlogCase StudiesROI CalculatorKnowledge BasePlatform ComparisonFind Alternatives

Company

AboutContactCareers

© 2026 Agxntsix. All rights reserved.

Privacy Policy|Terms of Service|Cookie Settings

Claude and Anthropic are trademarks of Anthropic, PBC. Agxntsix is an independent company and a member of the Anthropic Partnership Program.