Routed

Which city, which neighbourhood, which room: an agent on Qloo's taste graph routes indie tours. On tours announced after its AI's release, adding Qloo to city size found 13.2 pts more real cities.

Comment

Try it: Plan a tour · A finished tour book: Sierra Ferrell · Proof: Qloo on real tours · The MCP server · Code on GitHub

In 30 seconds

  • What it does: for an artist's tour, Routed answers three questions from Qloo's taste graph: which cities their fans over-index in, which neighbourhood inside each city, and which room there fits the crowd. Then it routes the dates and drafts a checked hold request for every venue.
  • What we measured: on 13 tours announced after the AI model's release, the city ranking Routed ships (Qloo + population) found 13.2 pts more of the real cities than population alone, better for 11 artists and worse for none (p < 0.001). On 49 older tours it found 7.4 pts more (36 better, 7 worse, p < 0.001).
  • What didn't go our way: on older tours the model alone does better, and Qloo's room and opener picks trail the model's. Both are under "How we checked".

Inspiration

Independent artists route tours on instinct and a streaming dashboard that counts plays, not people who'd buy a ticket. A language model can recite past tours; it can't tell you where this artist's fans over-index today, which rooms they use, which blocks they live on, or who else they'd turn up for. That is what Qloo's taste graph holds, so we built a booking agent whose evidence is Qloo and whose rules are a real booker's.

What it does

Name an artist, a starting city, a first date, the number of shows and the usual crowd. Routed:

  • asks Qloo where the fans over-index in the region: one heatmap of thousands of tiles (about 6,000 for North America), ranked into touring cities on one scale;
  • runs an agent (NVIDIA Nemotron 3 Ultra on Nebius Token Factory) that shortlists cities in Qloo + population order, scouts each one through Qloo for the rooms these fans go to and the artists who share their audience, then books a room per city and one support act for the run;
  • runs check_plan over the booking: the rules object (radius clause, major markets, no leg over 1,800 km with ferries and borders counted in hours, room sizes) and the agent re-books, up to two rounds;
  • measures every room (Wikidata first, then a web page that names it, source linked); a room "fits" only at 85–130% of the crowd, and parks are never rooms;
  • maps each city at street level from Qloo's city heatmap, names the fans' neighbourhood by a vote of the places nearest the hottest blocks, says whether the room sits in it, and puts the poster run there;
  • writes a hold request for every venue, with every figure checked against that stop's evidence;
  • shows the stops Qloo changed against the same rules run on city size alone, and prints the tour poster from the data.

Without Qloo, the same model sent Sierra Ferrell to Toronto, #202 on her fan map, and its cities averaged #50; Routed's averaged #10, with Asheville, her #1 fan city and only the 238th biggest, on the tour. Routed is also an MCP server, so any agent can call fan_map, fan_neighbourhoods, fan_profile, route_tour and get_tour.

A day sheet: the room sized to the crowd with its source, the fans' blocks at street level, the support act, and a checked hold request.

What makes it Qloo-powered

Every city, neighbourhood, room, support act and poster run starts from Qloo. A typical tour makes about 70 Qloo requests, each listed in its tour book with the exact request and the answer. The five that matter most:

  • /v2/insights with filter.type=urn:heatmap and one WKT polygon per region: where the fans over-index, on one scale;
  • the same heatmap with filter.location=POINT(city) and a radius: the city at street level, in tiles about 150 m across;
  • /v2/insights with urn:entity:place, filter.tags chosen by the crowd's size tier, and a point and radius: the rooms these fans go to, and the record stores and cafés near their hottest blocks;
  • /v2/insights with urn:entity:artist, signal.location and filter.popularity.max and .min: support acts clearly smaller than the headliner;
  • /v2/analysis/compare between the headliner and the support act: what the two audiences share, for the pitch.

How we built it

  • Next.js 15 on Vercel; Google Cloud Storage, keyless through workload identity federation. Qloo over REST through one request gate with daily and hourly allowances; at the cap, a visitor gets a stored tour instead of an error.
  • Nemotron 3 Ultra with strict JSON and reasoning off; deterministic rules decide everything that must be exact. 174 unit tests, and CI on every push.

How Routed is built: Qloo supplies the evidence, Nemotron makes the choices, rules check them and send objections back.

Challenges we ran into

  • Heatmap percentiles are relative to each response, so asking for the US and Canada separately floated Calgary to #1 for Tyler Childers. One polygon per region fixed it.
  • Qloo tags a 150-cap bar, a county fair, a park and a 19,000-cap arena all as live music venues. Rooms are asked for by size tier, then filtered and measured.
  • A booking agent's review found the band's own singer as its opener, no London on a UK tour, 4,400 km jumps and a park for 15,000. Each is now a rule with a test.
  • In a free tool loop, Nemotron repeated lookups until it ran out of turns. Now the moves are fixed and the choices are the model's.

Accomplishments that we're proud of

  • Qloo measured, not asserted: three tests on real tours, each committed before it ran and published whatever it said.
  • Rules a booker would sign, logged when they override the agent: London stays, no leg over 1,800 km, rooms sized to the crowd.
  • A poster an artist would put up, and a street-level map of where their fans are in every city.

Qloo finds, the agent decides, check_plan objects and the agent re-books. Every Qloo call is listed in the tour book.

How we checked

All of it is on /proof, artist by artist.

  • Tours the model can't have read: the 13 tours announced after Nemotron 3 Ultra's release, from all 185 pages in Wikipedia's 2026 and 2027 tour categories. Qloo + population found 54.5% of the announced cities, population 41.3% (11–0–2, p < 0.001). Qloo against the model alone: −4.6 pts, not significant (4–8–1, p = 0.39).
  • 49 older real tours, 30 of them held out: Qloo + population beat population by 7.4 pts (36–7–6, p < 0.001); +9.3 pts held-out, p < 0.001; +4.5 pts benchmark, p = 0.06. Here the model alone is 12.7 pts ahead of Qloo's ranking (36 artists to 7). It has likely read these tours' announcements; after its release, the gap is 4.6 pts and not significant.
  • Inside the agent, swapping Qloo's table for the biggest cities moved the share of its cities on the real tour from 49% to 46%, within the noise (p = 1.00). It ran before the blend shipped.
  • Rooms and openers: the model alone wins. The real room was in the top 3 of Qloo's room ranking in 29 of 102 cities, and of the model's in 75; Qloo's opener matched 1 of 22 tours, the model's 6. These are 2023–2025 tours the model may have read about, and it knows each city's main rooms. We don't claim Qloo picks the room or the opener better.
  • On Qloo's own fan map, Routed's cities averaged #8.3 across twenty artists, the model alone #26.2. That is Qloo's own ruler, so it shows how far a model strays from where Qloo puts the fans, not that Qloo is right.

Known limits

  • A past tour is one booker's answer, and 13 post-cutoff tours is a small sample.
  • About a quarter of booked rooms have no confirmed capacity; the sheet says to ask the room. No avails, guarantees or ticket prices, and festival weeks aren't checked.
  • Qloo affinity says where fans over-index, not how many tickets will sell, and no working booker has validated Routed yet.

What we learned

Adding Qloo to a big-city list finds more of the cities artists really play, and on unread tours that blend is level with the model (+3.6 pts, not significant). The model knows rooms and openers better, so Routed measures rooms instead of trusting any ranking. Qloo goes down to the block, which turns "which city" into "which neighbourhood, and is the room there".

The same audience across the rest of Qloo's graph: podcasts, brands, mood, films, TV and books.

Who it's for, and what's next

Managers and small agencies who route without a data team (about $29 a month a seat), and promoters who want to know whose fans are dense in their city. No paying users yet. Next: five working managers using it, with their edits published.

Built With

Share this project:

Updates

Submission history