How LinkedIn Unified AI Products Around a Target Audience Graph
Understand how a shared "target audience" representation solves cold-start, decay, and thin-market problems in recommendation systems, and why unifying AI infrastructure often requires unifying teams first.
4 sections ยท 9 lessons
Course outline
The Fragmentation Problem
- Why fragmented products often mirror fragmented org charts
- The cost of shipping parallel AI features instead of shared foundations
Building a Target Audience Graph
- Cold start: inferring audience with LLMs
- Warm signals: time-decayed activity weighting
- Thin markets: fusing buyer-side and seller-side signals
AI Features Built on a Shared Foundation
- Embedding retrieval + multi-stage ranking
- Two-stage LLM pipelines for cost-efficient scale
The Organizational Lesson
- Design around the customer problem, not the org chart
- Why reusable foundations compound over future products