SuiteScout — Resort PMS sales-ops

SuiteScout finds resorts that need ViveReply's Resort PMS and drafts the first message. You approve, it sends — live research, a deterministic 0–100 score, and a human-approved draft for every prospect.

Live demoGitHub0

Overview

SuiteScout is ViveReply's Resort PMS sales agent — an AI sales-ops tool that finds resorts that need ViveReply's Resort PMS and drafts the first message to them. It was built for the Vultr Agent Rush hackathon (Track 02 — Powering the Future of Work, Challenge 2: Sales Operations) and is live at suitescout.vivereply.com.

The problem

Most resorts in Bangladesh still take bookings by phone, Facebook message, or through online travel agencies that take 15–20% of every booking. They lose direct bookings because they have no website, no booking engine, or no system at all — and the company selling them the fix, ViveReply, had to find those resorts and write the first message by hand, one at a time.

What it does

SuiteScout does that first mile. Its five-step loop — research, score, draft, log, follow-up — runs on every prospect. Research fetches the resort's live public website and listings and extracts verifiable facts. Scoring is a deterministic 0–100 rubric in code: no website +30, no booking engine +20, phone-only +15, OTA-dependent +15, demand signals +10, scale +10; direct booking −40, an existing PMS −30. Bands are Hot 70+, Warm 40–69, Cold below 40. Drafting is done by an LLM, but every draft carries at least one genuine observation about that resort and never contains prices. Discovery starts from OpenStreetMap area search — a phone-only or Facebook-only resort is a good lead, not a bad one — plus manual adds.

Nothing goes out without a human. Every outreach send goes out only from an approved draft. Sending is capped at 10 emails a day, one email per prospect in total, with List-Unsubscribe on every message and auto-send off by default.

Under the hood

A Next.js 15 + TypeScript app in Docker on a Vultr Cloud Compute VM in Mumbai, with a shared self-hosted PostgreSQL 16, Caddy TLS, and nightly backups. AI drafting runs on Vultr Serverless Inference (DeepSeek V4.1 Flash; Qwen 3.8 Flash Next is selectable). The demo video walks through a live run on Balishira Resort: https://youtu.be/4V5lu60qQC8

Honest limits

Stated in the README: there is no scheduled sending yet and no suppression list yet — a STOP reply is not yet recorded automatically. Those are the first two items on the post-submission build list, ahead of deal stages and a follow-up engine.

How it uses Vultr

Vultr Cloud Compute instance for hosting the app and self-hosted PostgreSQL. Vultr Serverless Inference for scoring in code and LLM drafting.

No comments yet

Be the first to comment on this project.