ELATUM

The problem

Short-term rental owners in the Smokies handle dozens of repeat guest questions every week — WiFi passwords, hot tub temps, check-out instructions, bear-trash rules, Cades Cove hours. After 11 PM, every unanswered text becomes a slow response — which Airbnb penalizes in search ranking and Superhost status.

Owners didn't need another hire. They needed something that runs the playbook 24/7, sounds like them, and — critically — actually understands the local area. Generic AI can't tell a guest whether Cades Cove is vehicle-free on Tuesday or when Elkmont fireflies are running. ELATUM can.

The approach

ELATUM is a 7-node AI pipeline purpose-built for guest support, not a generic chatbot pointed at a knowledge base.

1. Ingest — guest message arrives via Twilio SMS or web chat
2. Guest lookup — phone matched to active reservation
3. Intent classification — sorted into 8 categories
4. RAG retrieval — semantic search over property + local docs (PostgreSQL + pgvector)
5. Response generation — grounded in retrieved context, no fabrication
6. Action execution — reply, Stripe link, maintenance ticket, or owner escalation
7. Audit log — every action visible in the host dashboard

The moat: local data

The ELATUM knowledge base isn't web-scraped. Every property gets a bespoke RAG corpus built from the owner's own documents, layered over a shared East Tennessee knowledge layer that maintains things generic models can't:

• Cades Cove vehicle-free day schedule
• Bear-proof trash procedures by municipality (Townsend / Walland / Pigeon Forge)
• GSMNP trail closures and alerts
• Elkmont synchronous firefly season
• Local dining, seasonal events, and Maryville College / Highland Games traffic

Modular by design — the Petals framework

ELATUM ships as independent, plug-and-play modules. V1 launches with one Petal; each subsequent module extends the platform without a rebuild.

• The Concierge (V1, 2026): 24/7 guest support, the current shipping product
• The Yield (planned): dynamic pricing using local demand signals
• The Guardian (planned): STR compliance monitoring (noise, fire, trash)
• The Pulse (planned): internal ops and maintenance routing

The stack

Deliberate and battle-tested. One codebase, one database, one LLM provider.

• FastAPI (Python) backend
• LangGraph for agent orchestration
• Specialized LLM for intent classification and response generation
• PostgreSQL + pgvector for state and semantic search
• AWS (EC2 + RDS) for production
• Twilio for SMS, Stripe for payments, Guesty PMS integration in progress

Status

ELATUM is in beta, launching May–June 2026. Current focus: onboarding the first cohort of Smokies-area operators and tuning the agent against real guest traffic. Metrics and named customers will land here as the beta matures.

If you run a short-term rental in the Smokies and want into the early cohort, the product site is the place to start — or try the live demo and book a 15-minute walkthrough.

What this build proves

Two things. One: a single experienced builder can take an AI agent from idea to shipped product without an agency-of-five overhead. Two: niche agents with real local knowledge will beat generalist tools that understand all industries shallowly. That's the bet the agency is built on — and ELATUM is the proof of concept.