CreatorMatch

A Moldova-only web platform that matches businesses with local Instagram, TikTok, and YouTube creators through an AI-ranked shortlist.

Role

Solo: design and build

Frontend

Next.js, React, TypeScript

Backend

FastAPI, Postgres + pgvector, Redis, Celery

AI

Claude, vector embeddings

Overview

Global influencer tools (HypeAuditor, Upfluence, Modash) have thin-to-zero coverage of Moldovan nano and micro creators, so local businesses have no real way to find them. CreatorMatch closes that gap with a curated local database those tools ignore. A business enters a product and a budget; the platform returns a ranked shortlist of Moldovan creators, each with a fit score, a known or estimated price, a score breakdown, and a ready-to-copy English outreach draft. It is scoped to discovery: it produces the shortlist and the drafts, and the business runs the actual outreach.

What I built

A multi-tenant web app on Next.js and FastAPI, structured as four independently testable cores: a creator repository (Postgres + pgvector), a pure-function pricing estimator, an explainable ranking engine, and an LLM service behind a single swappable interface. Ranking runs in two stages to keep LLM cost bounded: pgvector cosine similarity retrieves candidates over the whole pool cheaply, then Claude scores brand fit and drafts outreach for only the top candidates. A weighted, tunable score (40% fit, 20% reach, 20% price, 20% quality) orders the results, with every sub-score exposed as a per-result breakdown. Celery workers handle database seeding, scraping, and embedding backfill off the request path, and the pricing estimator prices every creator from Moldova-localized CPM benchmarks when no real rate is known.

Highlights

• Two-stage matching: cheap pgvector retrieval over the full pool, then LLM re-scoring on the top candidates to control cost.
• Explainable ranking: a tunable weighted formula (fit / reach / price / quality) with a visible breakdown per creator.
• Reach-based pricing estimator with Moldova-localized CPM benchmarks, so every creator carries a known or clearly-labelled estimated price.
• A curated local creator database as the moat, kept fresh by background scraping and manual curation.
• Multi-tenant business accounts, LLM-drafted English outreach, and CSV export of results.

Gallery