Applied AI systems / Local prototype / Engineering notes
Skina
A fictional retail advisor that selects a basket using deterministic rules and explores gated model explanations.
React / TypeScript / Node.js
Local fictional-fixture tests and a frozen evaluation include preserved misses. Gated adapter observations do not establish general model reliability.
The work
A fictional retail advisor that selects a basket using deterministic rules and explores gated model explanations.
Evidence
Local fictional-fixture tests and a frozen evaluation include preserved misses. Gated adapter observations do not establish general model reliability.
Scope
Latest retailer prototype is local. Its shopper interface uses mocked explanations; the older public preview is separate. Real-user usefulness and live AI improvement remain unproven.
The decision, at a glance
Who chooses the basket?
- 01
Review constraints
Fictional products, budget, owned categories and exclusions.
- 02
Select deterministically
Category-order rules select the basket and calculate totals.
- 03
Explain within bounds
Mocked UI explanations; a separately gated adapter cannot alter the selection.
Frozen fictional evaluation preserves misses. Limited adapter observations do not establish real-user usefulness.
Read the verification scope ↗Original explanatory diagram. Local prototype; the older public simulation is separate.
The problem
Explore how a fictional retail advisor can explain a product shortlist while respecting shopping preferences and budget.
Implementation
I built a local React and TypeScript flow with a fictional catalog. Deterministic rules select eligible products in category order using budget, owned items, texture and exclusions. Selection is greedy, not an independently optimized basket.
One decision: separate selection from explanation
The application chooses products and totals. A gated model adapter can explain selected records but cannot change the basket, prices or totals. The local shopper interface uses explicitly mocked explanations; the live adapter is not wired into it.
Demonstration
The local retailer prototype uses fictional products and synthetic inputs. An older public simulation-first preview is separate and does not demonstrate this retailer flow.
Evidence and limitations
Local fixture tests and frozen evaluation preserve misses. Limited gated adapter observations do not prove general reliability, improved usefulness, clinical validity or real-user outcomes. No real retailer or product catalog is represented.