Multi-Model Pipelines

For complex apps, one call isn't enough. You build a Chain.

The Filter-Writer Pattern

1. Step 1 (Filter): Use Gemini Nano/Flash. "Is this user input safe and relevant to our product? Answer YES/NO."
2. Step 2 (Logic): If YES, pass to Gemini Pro. "Answer the user's question in detail."

def safe_answer(user_message):
    # Cheap check
    if call_flash_guardrail(user_message) == "NO":
        return "I cannot answer that."
        
    # Expensive Answer
    return call_pro_model(user_message)

The Summary-Expansion Pattern

1. Step 1: Feed 10 documents to Flash -> Get 10 summaries.
2. Step 2: Feed 10 summaries to Pro -> Write a final executive report.

Summary

Chaining optimizing cost (using cheap models for bulk work) and quality (using smart models for the final polish).

In the final lesson of this module, we handle Errors.