Semantic Ranking with AI

[alexphiev]

Description:
Use LLM to rank filtered candidates based on user context.

Tasks:

• Create prompt template: "User wants: [query]. Rank these places: [candidates with summaries]. Return JSON with place_id and relevance_score."
• Call Gemini API with filtered candidates (1 call per search)
• Parse JSON response and re-order results
• Cache rankings for identical queries
• Add timeout and fallback to score-based ranking

Acceptance Criteria:

• Ranked results feel personalized to user query
• Handles edge cases (API failure, timeout)
• Cost: <€0.01 per search