LLM (Large Language Model)

In a product, the LLM is rarely used as a standalone piece. A sales email drafting tool, for instance, combines an LLM with CRM data (via RAG) to personalize each message with the customer's actual history. The model choice affects cost, latency, and quality, but it's almost always the integration around the model (data, prompts, controls) that determines whether the final product actually works in production.

A common mistake is defaulting to the most powerful model on the market without comparing cost and latency against the actual need. For a simple, high-volume task (classification, extraction), a smaller, cheaper model often performs just as well at a fraction of the cost. The right approach is to test several models on the specific use case before locking in a choice.