Research topic
Personalization and decision support
Personalization across regions and languages, kept fresh, safe and running in production, and agents that support decisions in small firms.
Each paper states its method, puts its evidence in captioned tables and ends with its limitations.
Papers
Event-driven decision making in commerce: a design for interpreting signals as they arrive
We describe an event-driven design that turns commerce signals into merchandising, pricing, payment and delivery decisions under explicit constraints.
Agent-based decision support in commerce platforms for small and medium-sized firms: a conceptual framework
We propose a conceptual framework for agent-based decision support built into commerce platforms used by small and medium-sized firms.
Detecting and responding to distribution shift in a deployed personalization system
We describe a design for detecting covariate, label and concept shift per cohort in a personalization system, and for containing it with fallbacks.
Operating a personalization system across markets: launch, experimentation, rollout and incident response
We describe the operating procedure for a multi-market personalization system: how models are launched, tested, ramped, retrained and recovered.
Distributed inference for personalization across regions: a serving design
We describe a serving design for personalization across regions: residency-bound cells, two-stage ranking, tail-tolerant requests and a fallback ladder.
Governance and safety controls for a multi-region personalization system
We describe governance and safety controls for a personalization system, from lawful basis at ingest to model documentation, fairness checks and audit.
Feature freshness as a pipeline property: a design for time-sensitive personalization features
We describe a feature pipeline that treats freshness as a measured, budgeted property of every feature, with decay, partial refresh and stale-feature fallback.
Multilingual embeddings for cross-region recommendations: a design for one item space across languages
We describe a design that maps catalog text in many languages into one embedding space, so recommendations can transfer across regional markets.