Case Study · AI Product
Album Art Moderation
Vision AI · Policy · Human Review
Every album cover uploaded to a music platform ships straight to listeners. You can't review them all by hand, and you can't auto-approve the ambiguous ones. I built a two-stage pipeline that spends money only where the answer is actually in doubt.
The Problem
Album artwork is user-generated content with no gap between upload and audience. A platform can't put a human in front of every cover, and it can't send every cover to an expensive model either — most are obviously fine. But the two ways of being wrong don't cost the same. A false approval puts harmful content in front of users; a false rejection just adds an image to a review queue and annoys one artist.
Users & What I Learned
The human reviewer
Doesn't want a verdict handed down with no reasoning. Wants to see what the model saw, which policy line it applied, and what the argument on each side was.
Critical insight: the gray zone is the product
All of the cost, all of the risk and all of the design work live in the band where the model isn't sure — so that band is where I put the expensive machinery.
Decisions I Made and Why
Two stages, not one
A fast vision pass scores each cover across six policy categories; clear cases exit immediately and cost nothing. Only the middle band reaches the second agent.
Approve at 10%
The threshold is a risk-tolerance dial. A 10% approve threshold means requiring 90%-plus confidence that a cover is clean before it passes unattended. Tighter costs more; that's the trade.
Deliberation design
The second agent reads the policy document, then works through fixed steps — observation, policy check, the case for approval, the case for rejection — before it rules.
Processing Pipeline
Image Upload
Artwork is submitted to the portal.
Scan
Agent 1 scores all six policy categories and returns a confidence score.
Deliberate
Only gray-zone images reach Agent 2, which argues both sides using the policy document.
Decide
Final ruling: approve, reject, or escalate to human review with full reasoning.
Pipeline: Image uploaded → fast scan scores policy categories → clear cases exit → gray-zone cases deliberated → unresolved items escalated to human review.