The problem Prenew sells refurbished gaming PCs across Europe. Finding creators to work with is still manual: scrolling YouTube, TikTok and Instagram, checking follower counts and guessing whether the audience fits. Prenew told us existing tools missed many creators, especially in small markets, and couldn't be customized. Small creators and small markets, often the best value, are the easiest to miss.
What we built Prenew Creator Scout automates the search and explains every result. - A discovery engine in Python running on a small Google Cloud VM, with a Twitch poller. - A triage app (Lovable) on a Google Sheet via Apps Script: review creators with the arrow keys, graded in Prenew's own gold / silver / bronze tiers ("we grade creators the way you grade PCs"); filter by market, size, game and contact; ask questions in plain language ("reachable Czech tech creators under 50k"); and "Find more creators like your good fits", which runs discovery again from the marketer's own picks. - Every creator is one row with Prenew's minimum fields: country, subscribers, average views (last 30 days, or 90 for less active channels), niche and game, and contact or agency. It downloads as a CSV if the Sheet isn't needed. - Prenew's past partners are flagged, so known creators are never presented as new finds.
How it works (and what's AI about it) - Starts from a few seed creators (Prenew's past partners) or from nothing: a web search picks local seed channels automatically, so a new country works on day one. - Expands on YouTube through featured channels, mentions and the people who comment on the seeds' videos, then keeps active channels with 1,000+ subscribers and no upper cap. - Reads each creator's own links for Instagram, TikTok, email, agency and affiliate or shop links, and flags creators who invite business inquiries. - AI: Claude labels sampled audience comments (hardware question, purchase intent, genuine, spam), rates brand fit, names the niche and main game, and writes brand-safety notes (for example very young audiences). In the app, AI turns a plain-language question into filters only; every result is a real row. - The language of the comments shows how local the audience is. Local reach (views x local audience share) shows when a big foreign channel reaches more locals than a small local one. - Adds Twitch streamers (live polling), podcasts (Podcast Index) and TikTok creators (web search plus TikTok's official oEmbed captions), merged into one row per creator. A TikTok-first flag catches creators whose small YouTube channel mirrors a much bigger TikTok (about half of Prenew's partners). - Scores fit transparently: every factor is its own column, with two real comments as evidence. No black box, and the weights can be changed.
Example: a Finnish Minecraft channel with only 3,000 subscribers averages 23,000 views per video over the last 30 days, and 96% of its comments are in Finnish. Follower-count tools rank it low; Scout ranks it first in Finland, with a brand-safety note that its audience is very young.
Results in one weekend: 1,898 creators across 11 countries (Finland, Germany + Austria, Poland, Czechia, Slovakia, Slovenia, Estonia, Latvia, Lithuania and Romania), including 124 YouTube channels under 10,000 subscribers. A market run costs about $1 in APIs. The tool found 6 of 9 Finnish partners we held back from it, and its automatic seed search surfaced 4 of Prenew's Baltic partners without being told who they were.
How it differs: Modash-style databases cover public profiles but can't tell whether a creator's audience asks about PC hardware or speaks the local language. Creator platforms like Boksi only know creators who signed up. Scout finds the ones who didn't and checks Prenew-specific fit.
Limitations today: recall is weaker where seeds were narrow (0 of 3 Polish partners on the first run); TikTok follower counts are typed in by hand because we never scrape TikTok; Instagram is reached only through creators' own links.
Responsible by design: official APIs only, no scraping of Instagram, TikTok or Spotify, comments used in aggregate, company and media accounts filtered out, public business contacts only.
What's next: move from Sheets and a VM to Cloud Run plus Postgres with Google sign-in (about $100-150/month for 11 markets refreshed weekly), add Kick streams and creator-agency rosters as sources, Instagram's official Business Discovery API, a weekly digest of new top creators, and a hand-off of good fits to Prenew's CRM.
Team: Diana Băcîrcea and Claude Code. The code is in a private GitHub repo; access on request.
