Problem
Streaming alone is hard, and it's easy to skip a day. With a guest it's easy to fill an hour, but 34% of collaboration attempts never became a conversation.
StreamYard
Guest matching for new streamers.
The solution
Led a small squad building Collab, a two-sided marketplace where creators find guests for their streams and guest spots on other streams. Also led product and engineering for the Teams/Business tier.
Approach
Replaced category-based discovery with vector similarity, then rebalanced scoring toward content similarity over raw follower count: recommendation acceptance up 25%
Paired newer creators with slightly larger ones: motivating, and good for their growth
Built an Opportunities board where hosts post guest requests, like a job posting, and creators apply: 58% of new conversations
Cut first contact to one click and added AI conversation starters: conversation initiation up 32%
Launched invite-only in cohorts to balance supply and demand per niche, hand-curating the first 200 profiles
Ran human moderation alongside automated screening, plus blocking and reporting: 40% fewer policy violations, 60% faster approvals
For the Business tier, introduced a workspace abstraction and numeric RBAC, migrated in four independently-deployable phases with zero downtime
Next case
Whistle
Dance teachers needed video that showed movement clearly. Built live classes for full-body instruction and uneven networks.