Influencer Matcher

Status: complete · Flask · fuzzy matching · LiteLLM · Notion API

The problem #

For a testimonial campaign, over a hundred collaboration partners each needed the right product assigned to them. The basis for it — years of collaboration history — existed, but only as a mess of spreadsheets with inconsistent names and handles.

What I built #

The tool searches that history fuzzily, so it copes with divergent spellings and handles, suggests one product per person, and has a language model justify the match. Profiles, ratings and notes sync with Notion.

The model integration is built provider-independently, and stored credentials are encrypted at rest.

Why it is built this way #

Fuzzy matching does the work that has a right answer; the language model does the part that needs judgement — explaining why this pairing makes sense. Asking a model to do the lookup as well would have made the result unverifiable.