Echolocate

About

Echolocate finds sponsors for small events (campus shows, community gigs) by asking Qloo which brands and local businesses the event's crowd already likes, then drafts the pitch email and a one-page sponsor sheet.

Most events are small. In 2024, Eventbrite creators alone hosted nearly 5 million events and issued 270 million tickets (Eventbrite Form 10-K), about 55 tickets per event. Big tours have sponsorship teams; events this size don't, and they have no sponsor history to copy.

Live demo: https://echolocate-topaz.vercel.app/ · Demo video (1 min): https://youtu.be/S9hfe2OjMYE

Try it in 60 seconds (no sign-up)

1. Click Try an example: a 150-person 21+ bar gig in Athens, GA with local bands Qloo doesn't know, so it searches by genre. Local sponsors come first, on a map. The link under it runs a bigger campus show in Austin, TX (800 people, with Phoebe Bridgers, boygenius and Clairo), where the reasons name the artists. Both are answered instantly from saved, real Qloo answers.
2. Scan the brands by category: each has a Sponsor Fit Score and the reason from Qloo's data.
3. Click Draft pitch on any brand or local place, then Make one-pager and Download PDF.
4. Scroll to How the agent got this: the decisions the agent took, and every Qloo request with its parameters.

Built for the Qloo Agentic Hackathon.

What it does

The organizer enters the kind of music (or artists on the line-up), the city, who mostly comes, who is allowed in (all ages, 18+ or 21+) and the expected crowd. Echolocate then:

1. ranks brands per category for that crowd, each with a Sponsor Fit Score and a plain reason,
2. shows local cafes, clothing stores and (for 21+ events) bars on a map,
3. drafts a pitch email for any brand or local place,
4. makes a one-page sponsor PDF in the browser.

Why this is an agent

Given one goal, "find sponsors for this event", Echolocate plans and runs the Qloo calls itself, adapts when an answer is empty or thin, stops to ask when a choice would change the result, checks the language model's output, and hands back work the organizer can use: a ranked list, an email and a PDF. It never sends anything; the organizer edits and sends.

• Resolves vague input into Qloo IDs, and asks when unsure. Artist names go to /search; only an exact name match is used. Anything else is shown back as candidates to pick from, because a search for "Wishy" returns "Wish You Were Here". A genre not on our checked list is looked up with /v2/tags, and the page says which Qloo tag it became.
• Adapts to the answers. If Qloo has no taste data for the city, it asks again for the line-up alone and says so. If fewer than 6 local places come back, it widens the search radius once. For a crowd of 200 or fewer it puts local businesses first.
• Applies fixed rules the model cannot override. Alcohol brands and bars only for 21+ events; advocacy groups and institutions are removed. All on the server.
• Checks the language model. The model only writes the email body, from a fixed list of facts. A plain code check throws the draft away for a template if it adds a number that is not in the facts, calls a local place a "venue", or claims a "perfect fit", and the page says why.
• Shows its work. Under every result, "How the agent got this" lists what it decided for that search and each Qloo request with its parameters and result count.

How the agent works

Every Qloo call is listed in the app under "How the agent got this", with its result count.

1. Genre. A genre picked from the list already has its Qloo tag id (src/data/music-genres.json), so no call is needed. A typed genre is looked up with GET /v2/tags (filter.query), and the app says which Qloo genre it matched.
2. Artists (optional). Each name goes to GET /search (query, types=urn:entity:artist). Only an exact name match is used. Otherwise the app asks the organizer to pick the right artist, because a search for "Wishy" can return "Wish You Were Here".
3. Brands. GET /v2/insights with:
   • filter.type=urn:entity:brand
   • signal.interests.entities (artist ids) and signal.interests.tags (genre tag)
   • signal.location.query (the city) and signal.demographics.age
   • feature.explainability=true

여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗