Data & AnalyticsClassificationTrust & SafetyFixed budget
LLM Moderation Pipeline for a Marketplace
Haven Trust & Safety · Remote (Global)
$10k fixed
The outcome
An LLM classifier that reads listings and messages, catches policy violations with nuance, and routes the gray-area cases to humans.
The client
Haven Trust & Safety — Remote (Global). A data & analytics build we delivered end to end.
The challenge
Build a content-moderation pipeline that scores listings and messages against policy, auto-actions clear cases, and queues ambiguous ones with an explanation.
What we built
- Design a policy-aware classification pipeline with an LLM
- Calibrate thresholds against a labeled eval set
- Route uncertain cases to human review with rationale
The stack
Stack: LLM classification, eval/calibration, and a queueing integration.
Want something like this?
Tell us what you're building and we'll scope it — most projects start within a week.