The Scenario

Your AI startup is burning $5,000/month on OpenAI bills. Users ask the same questions in slightly different ways:

• "What is your pricing?"
• "How much does it cost?"
• "Price list please"

Exact match caching fails here. You pay for all 3 queries.

The Goal

Implement Semantic Caching using vector similarity.

1. Convert the user query to an embedding (vector).
2. Compare it with cached query embeddings.
3. If similarity > 0.9, return the cached response.

Helpers Provided:

• mock_embed(text): Returns a list of floats (vector).
• cosine_similarity(v1, v2): Returns a float between 0.0 and 1.0.

Requirements:

• Iterate through the cache.
• Find the cached query with the highest similarity.
• If similarity > 0.9, return cached response.
• Otherwise, call LLM and cache the new query/response.