LG Electronics Applied AI recommendations
An unavailable product should not end the journey.
A catalog-driven recommendation pipeline that finds suitable in-stock TV alternatives and explains each suggestion, combining business constraints with AI-assisted ranking.
Designed and built the recommendation pipeline, its quality checks, and the evaluation approach.
Operational pipeline · customer rollout pending
What is a useful alternative, and why?
Ranked product alternatives with human-readable reasons, prepared for consistent use across customer communications and commerce workflows.
Conceptual workflowThe business problem
When a product is unavailable, a customer needs a credible alternative. Teams need consistent recommendations that respect product suitability and availability, with explanations that make the choice understandable.
What I built
Built a hybrid recommendation workflow that narrows the catalog with availability and suitability rules before ranking alternatives.
Used product descriptions and attributes to compare candidates, then applied AI-assisted reranking and generated explanations grounded in catalog information.
Added automated publication checks and independent AI evaluation of recommendation quality, explanation accuracy, and labeling.
How I evaluated it
Checked eligibility and output integrity automatically, then used an independent AI reviewer to assess ranking, explanation quality, and labels. Evaluation identified messaging decisions to resolve before customer exposure.
Built to operate
A recurring catalog workflow with checks before publishing and fallback handling for model availability. The system can return fewer alternatives when the catalog does not support a suitable match.
What the work enables
Delivered an operational pipeline producing ranked alternatives and explanations for downstream use. Customer-facing messaging and rollout decisions remain separate from pipeline validation; sales impact has not yet been measured.