About HouseFull

A campaign agent for live music

Shows don’t sell out by shouting louder at the same fans. HouseFull uses Qloo’s taste graph to find the people next to them, then turns that into a plan a promoter, venue or artist team can run today.

What it does

From a booked show to a full room

Plan any show in New York, New Orleans or Los Angeles. In about a minute: new audiences, where they live, how to reach them, ad copy, partners, openers, a countdown and a show-night plan. Check in results and it re-plans.

Qloo-powered

Taste decides. Models only write.

Every decision comes from Qloo’s taste graph. Gemini names audiences, writes copy and checks it, but never invents who to target.

  • The next ring

    Taste-adjacent artists from the insights API, clustered into three new audiences

  • Where they live

    Per-audience heatmaps divided by a city baseline (lift), snapped to neighborhoods

  • How to reach them

    Podcasts, TV shows and brands each audience over-indexes on

  • Partners and openers

    Taste-matched places near the venue; smaller acts that bring the audiences

  • Outreach

    Contact details from Qloo entities: address, phone, website, hours

  • Tonight

    Affinity of tonight’s acts to the artists a fan likes; jazz clubs and pubs by tag

Lift, not popularity

Raw heatmap affinity tracks plain popularity (r ≈ 0.99): it mostly shows where people live. Divided by a city baseline it drops to r ≈ 0.07, real taste signal.

Every claim is sourced

Each Qloo call is logged with an id, and every reason in a brief carries the id of the call it came from. Unverifiable claims are dropped.

Architecture

How it’s built

Browser

Promoters, venues, artists, fans

AI assistants

8 tools over MCP

Web app

Next.js on Vercel

Clerk

Sign-in · organizations = workspaces

API

NestJS on Cloud Run · REST, live streaming, MCP

Agent

Pipeline · brief · critic · re-plan · ask · outreach

Qloo Taste AI

The taste graph

Gemini

Names, copy, critique

Ticketmaster

Shows tonight, conflicts

MTA

Subway alerts, late service

Postgres

PostGIS · pgRouting · runs & briefs

A run executes inside its live stream; an ordered event log lets any instance replay it

Responsible by design

What it will never do

  • Target traits

    Audiences are taste, never age, gender, ethnicity, religion or sexuality.

  • Rank by safety data

    Incident counts are show-night context for promoters only.

  • Track your location

    Tonight computes distances in your browser.

  • Make things up

    Numbers must come from the facts; a critic checks every brief.

Credits

Data, tools and license

Data

  • Qloo Taste AI (hackathon API)
  • Ticketmaster Discovery API
  • NYC Open Data (DCP, NYPD), data.nola.gov, LA GeoHub (CC BY 4.0)
  • MTA GTFS and service alerts
  • © OpenStreetMap contributors (ODbL), OpenFreeMap tiles

Built with

Next.js 16, React 19, Tailwind CSS 4, NestJS 12, Postgres with PostGIS and pgRouting on Supabase, Gemini on Vertex AI, Clerk, MapLibre. Hosted on Vercel and Google Cloud Run.

Open source under the MIT license →