Translating a product catalog with an LLM: cache keys and guard rails

A product catalog translation pipeline using LLMs often fails due to incorrect cache keys and lack of guard rails. This results in re-translated fields and incorrect descriptions. To fix this, use a cache key that captures every input that can change the output, and validators to treat the model's response as untrusted. Normalize input text to handle unstable whitespace and encoding differences. Use a unique key for each string produced by the pipeline, including source text, field, target language, model, and prompt version.