Module 5 Wrap-up: The Performance Guru

You have mastered the "Vibe" of the modern AI application. You know that Streaming is not just a technical feature—it's a requirement for high-quality user experiences. You also understand how to handle the "Traffic Jams" of production using Retries and Model Selection.

---

Hands-on Exercise: The Ticker Chat

1. The Goal

Create a Python script that asks Bedrock for a "Minute-by-minute summary of a fictional space mission."

2. The Implementation Plan

• Use converse_stream.
• Use flush=True in your print statement so the text appears instantly without waiting for a newline.
• Measure the Time to First Token and print it at the end.

3. Comparison

Switch the modelId to the largest Claude model (3.5 Sonnet) and then the smallest (3 Haiku). Note the massive difference in speed.

---

Module 5 Summary

• Streaming: Essential for low-latency perception.
• TTFT: The most important metric for chat UX.
• Throttling: Handling AWS limits gracefully.
• Model Tiers: Using smaller models for speed and larger for reasoning.

Coming Up Next...

In Module 6, we move from "Scripts" to "Services." We will learn how to wrap our streaming Bedrock calls into a professional FastAPI REST API so any frontend (React, Mobile) can talk to our AI brain.

---

Module 5 Checklist

• I can write a loop that iterates over a converse_stream.
• I know which block in the event stream contains the actual text.
• I understand the concept of Exponential Backoff for retries.
• I can describe why Haiku is faster than Sonnet.
• I have measured the latency of my first Bedrock stream.