SocialDB — the open database of the creator economy

5.8M creator profiles across 15 platforms. A marketplace where brands and creators do sponsorship deals directly, with escrow. Built and shipped by one person.

💡 What inspired it

Sponsorships are how creators actually get paid, but the market is absurdly inefficient: brands can't find the mid-tail creators who deliver the best ROI, creators without an agency get ghosted or give up 20–30% in commissions, and neither side trusts the other enough to send money first.

IMDB solved discovery for film. Crunchbase solved it for startups. Nobody had solved it for the creator economy — so I decided the database itself would be the product and the growth engine: index every creator publicly, let them claim their own listing, and put a deal marketplace on top.

🛠 How I built it

The core is a data machine: 15 import pipelines (YouTube, TikTok, Instagram, Twitch, Kick, podcasts, Substack, Bluesky, Mastodon, X, and more) that discover creators, refresh stats, and harvest public business contacts — 5.8M listings, 70K brands, and 1M+ verified contact emails, built almost entirely on free/public APIs instead of paid data vendors.

On top of that sits the marketplace: claimable listings (email-match auto-claim + bio-code verification), deal flow with escrow-protected wallet payments, AI-powered natural-language search ("gaming YouTubers with 1M+ subs in Canada"), an AI campaign agent, media-kit generation, and Pro subscriptions.

Everything runs on a single VPS — the entire company costs less than a gym membership per month. Discipline over budget: a resource governor pins every batch job to spare CPU cores at lowest priority so the website always wins, and a DB gate defers crons whenever live traffic needs headroom.

Every metric on a listing is computed from public signals — e.g. engagement rate is just

$$E = \frac{\bar{L} + \bar{C}}{F} \times 100$$

(average likes + comments per post over followers) — benchmarked against platform-and-size cohorts so a 50K-follower creator is compared to peers, not to MrBeast.

📚 What I learned

• Distribution can be a database. Our best acquisition channel is the index itself: "we already have your profile — come claim it" converts because the value is visible before signup. 13,700 creators have claimed listings; 12,000+ users signed up with $0 ad spend.
• At millions of rows, every lazy query is a outage. A single LOWER() around an indexed column made every login full-scan 7.6M rows (26 seconds under load). A missing index turned a 5s query into the cause of site-wide 522s. I now EXPLAIN first and measure everything.
• Free tiers are a strategy, not a hack — but they demand engineering: key rotation, quota-aware fallbacks, and zero-quota scraping paths for when the official API runs dry mid-day.

🔥 Challenges I faced

• Platform hostility: signed thumbnail URLs that expire, IP blocks, captcha walls, APIs that vanish. Each importer needed its own survival strategy — retries with backoff, residential relays, embed endpoints, and self-healing when creators rename their channels.
• Solo-scale operations: one night the server hit load 200 from stacked cron jobs; another day MariaDB OOM'd at 3 AM. The fixes became infrastructure: flock-guarded governed crons, swap provisioning, atomic deploys (a live user once caught a half-written file mid-scp — never again), and site-wide error tracking so nothing fails silently.
• Trust as a feature: escrow, ID verification, GDPR exports, listing takedowns, and platform-ToS compliance (e.g. no derived metrics on YouTube listings) — the unglamorous 40% of the codebase that makes strangers willing to transact.

📈 Where it is now

12,300+ registered users · ~8,600 monthly actives · 27 paying subscribers · 1,500+ deal conversations · first escrow deals completed end-to-end — and a 20-user interview program running right now to decide what gets built next.

Built With

• claude