Very simple UI, like the semantic search prototype
1 version: Predefined subset of English Wikipedia articles
Expected behavior:
User selects an input article from the pre-defined list (via dropdown, search bar with autocomplete, or something else)
User sees up to 3 question recommendations (left column) and 1 semantic search result for each recommended question (right columns)
Semantic search results should only display sections from the input article
Question recommendations are generated and ranked via an LLM prompt (we’ll figure out prompt details below within the Delivery Plan)
Question recommendations are filtered to exclude potentially harmful or inappropriate content (we’ll figure out exact filtering logic below under Delivery Plan)
Question recommendations are pre-generated, stored, and fetched at runtime
Input: Page title or article URL (from a predefined list)
Table of results should include up to 3 recommended queries (ranked by LLM), and only the top semantic search result for each recommended query (ranked by relevance)