Semantic cache lab

Try it → see it → read → next

**Cache near-duplicate queries** — skip LLM calls when similarity beats your threshold.

On paths: Builder · Student

Try tools: Code & developers

Try yourself

Playground

Semantic cache lab

Near-duplicate questions should cache-hit — novel ones call the LLM.

What are your refund hours?

Recap

What you just did

You classified cache hit vs fresh LLM call for paraphrased questions.

Read

Sketch

def answer(q: str):
    hit = cache.lookup(embed(q), threshold=0.92)
    if hit: return hit.text
    out = llm(q)
    cache.store(embed(q), out)
    return out