Provider
Intelligence
Engine

Reads a coordinator's incoming brief email and extracts structured requirements: capacity, city, dates, event type, vibe, and budget signals — no manual form to fill in.

Before searching, the agent recaps its understanding in plain language. The coordinator confirms or adjusts with a single click — "Confirmed / Let me adjust / Search anyway." No free-text required.

When the brief is ambiguous, the agent asks targeted questions — then proceeds. It states its remaining assumptions explicitly so coordinators can override if needed.

Design and population of the proprietary provider database for multiple cities and types of providers — including vibe taxonomy, capacity ranges, event type suitability, and agency history fields. At first the attention will be on hotels, restaurants and event venues.

Matches brief vibes ("industrial loft", "elegant rooftop", "modern and understated") against provider profiles using vector embeddings — beyond simple keyword search.

One-click drill-down on any shortlist result: full profile, capacity breakdown, photos, coordinator notes from past events, and source metadata. Basically everything you need to know about each venue that you can use in your proposal.

Pulls structured data — photos, ratings, opening hours, address — from Google Places API as the backbone for any provider in the database.

Integrates with Salesforce CRM to sync customer/provider data, opportunities, and sales activities.

Filters by capacity, location, provider type, and availability before surfacing any result. A venue that doesn't fit the hard requirements never appears in the shortlist.

When the PIL has insufficient coverage for a city or venue type, the agent automatically triggers live web search to supplement. Discovered results are queued for PIL ingestion after coordinator review.

Works for any destination from day one. Cities where the PIL has deep data return richer results; new cities rely on live search. The agent is transparent about the difference — it never pretends to know more than it does.

Coordinators can refine constraints after seeing the first shortlist ("too formal", "nothing with outdoor space", "give me more options") and trigger a new search — without starting over from the brief.

Reads QT/'s existing Outlook email archive via Microsoft Graph API — extracts provider mentions, past RFP data, and coordinator notes from years of institutional memory that currently lives nowhere structured.

To be properly scoped in order to identify the restraints of the mining.

Scrapes industry portals for provider information, updates, and news — ensuring the PIL has the latest data on available services and offerings. Think of portals like CVent Sourcing Network, Event Inc, fiylo, meeting e congressi, etc.

The agent subscribes to industry newsletters via a dedicated mailbox. On schedule, it reads new issues, extracts provider mentions, and ingests them into the PIL.

Monitors relevant Reddit communities for venue recommendations and event reports — capturing "hidden gem" discoveries that event planners share publicly but don't appear in any commercial database.

Runs continuously in the background — detects venue closures, discovers new providers, refreshes stale data. The PIL stays current without coordinator effort.

Monitors social media platforms for venue recommendations and event reports.

To be properly scoped in order to identify the limitations and their workarounds.

Structures QT/'s past RFPs as a searchable library: what was shortlisted, what the client chose, what was rejected, what was won, what was lost. Built from Outlook archive mining (TBC), manually compiled structured data and voluntary coordinator curation.

On every new brief, the agent retrieves semantically similar past cases and uses them as ranking context. "When QT/ last organised a gala dinner for 80 in Munich, these venues won and here's why" — fed directly into the shortlist reasoning.

Uses the historical case library as a golden test set — automatically runs the agent against known-good past briefs to measure quality and catch regressions as the system evolves.

A clean internal web interface where coordinators receive shortlists, review venue cards, and take action — no email back-and-forth, no switching tools. Built for speed and simplicity.

After reviewing a shortlist, coordinators can mark results as "selected / rejected" and add a note — under 30 seconds per provider. This is what makes the PIL compound in value over time.

Every shortlist shows a clear breakdown of how results were sourced: "7 from QT/'s knowledge base · 3 from live search." Coordinators always know how much to trust the output — no hidden uncertainty.

The agent only responds to provider sourcing requests. Off-topic inputs are refused cleanly. It never fabricates a venue — thin coverage is surfaced explicitly rather than filled with confident-sounding guesses.

A clean internal web interface for administrators to manage and monitor the agent's performance with metrics like usage statistics and costs.

Role-based access control ensures that only authorised users can access specific features and data within the admin dashboard.

Admins can create, edit, and delete user accounts, as well as assign roles and permissions.