← All work
02 · AI artist-manager

An AI artist-manager that matches for cents.

Ari helps creators find and reach the right brands. I designed the Outreach engine — the matching, the data model, and the economics — so it's cheap and explainable by design.

Founding Product ManagerAgentAri TechnologiesOct 2025 – NowAI / data platform
matching engine · one LLM call
creator (Modash)
beautygen-zmicroIN
→ shared taxonomy →
brand (Apollo)
skincaredtcpremiumIN
Score = .35·tag_overlap + .25·geo + .15·tier + .15·demo + .10·freshness
<100ms
match query · 2k-brand pool
~$0.01
marginal cost / creator

The best AI decision was where not to use AI — one cheap LLM call for the fuzzy part, then boring, fast, explainable SQL for the match.

The design

A shared taxonomy turns matching into a join.

Both sides — creators (from Modash) and brands (from Apollo) — map onto the same tag dictionary via a single stateless LLM call. Once they speak the same language, "find brands for this creator" is a scored join over a pre-cached pool, not an API storm. A transparent formula (not a black box) keeps every match explainable to creators and brands.

Key decisions

The architecture calls.

01

Pre-cache Apollo, query locally

Amortize a one-time brand pull across every creator — collapsing marginal cost to near-zero and matching to a local join.

02

One taxonomy, both sides

Makes matching a join, keeps it explainable, and lets one prompt serve creators and brands alike.

03

Rules before ML

A weighted score ships in weeks and generates the exact reply/close signals a learn-to-rank layer later needs.

Outcome

Fast, cheap, explainable by design.

<100ms
match query over a
~2,000-brand pool
~$0.01
marginal cost
per new creator
1
LLM prompt serving
creators & brands
Next

India's Got Latent & Samay's Live Tour

Open →