Series
The Memory and Personalization Model
How the system knows the person it serves: a five-layer context hierarchy that loads only what each query needs, a preference model that learns from behavior rather than demographics, and the forgetting, consent, and temporal architectures that keep the representation accurate, private, and current.
BMT-05.01
The Five Layers
The Mixture of Contexts organizes a person's context into five layers with domain-specific activation. The MoC Router selects the minimum context each query requires, cutting token …
BMT-05.02
How the System Learns You
P-RLHF replaces population preference models with per-person learning that starts from starter templates and overrides them through behavioral observation. By interaction 500, the …
BMT-05.03
What the System Forgets
The forgetting architecture applies domain-specific temporal decay so the system serves the person she is, not the person she was. Medications decay slowly. Shopping preferences …
BMT-05.04
Who You Are Is Not One Thing
The Intersectional Context Engine captures identity across twelve-plus dimensions with context-dependent salience. Dimensions interact rather than add, and the system learns which …
BMT-05.05
The Consent Architecture
Consent is a real-time gate, not a form signed once. Every data flow carries a five-parameter consent context evaluated at query time. Revocation propagates to external parties in …
BMT-05.06
The Person Over Time
Temporal intelligence tracks circadian patterns, longitudinal trends, and life events across every concierge domain. The system that sees the trajectory can flag the trend before …
Synthesis