Definition
Vector representations that incorporate a user's unique identity, preferences, or historical context into the embedding space to ensure retrieval (RAG) and reasoning (Agents) are personalized. This is typically achieved by concatenating user metadata to queries, using specialized adapter layers, or maintaining distinct vector subspaces for individual users to prevent generic response bias.
Distinct from 'Global Embeddings'; it's about shifting a query's location in latent space based on *who* is asking, not just *what* is being asked.
"A custom-tinted lens that makes certain documents in a library glow brighter based on the specific reader's interests."
Conceptual Overview
Vector representations that incorporate a user's unique identity, preferences, or historical context into the embedding space to ensure retrieval (RAG) and reasoning (Agents) are personalized. This is typically achieved by concatenating user metadata to queries, using specialized adapter layers, or maintaining distinct vector subspaces for individual users to prevent generic response bias.
Disambiguation
Distinct from 'Global Embeddings'; it's about shifting a query's location in latent space based on who is asking, not just what is being asked.
Visual Analog
A custom-tinted lens that makes certain documents in a library glow brighter based on the specific reader's interests.