AWS Bedrock Prompt Optimization Workshop

This hands-on workshop teaches you how to build production-grade generative AI applications with a focus on cost optimization, performance enhancement, and operational excellence.

Target Audience: AI/ML Developers, Software Engineers working with agentic systems, DevOps Engineers deploying GenAI workloads

---

Workshop Objectives

This workshop focuses on optimizing three key metrics for production GenAI applications:

Objective	Definition	Why It Matters
Accuracy	The quality and correctness of model outputs relative to expected results	Ensures your application delivers value to users and meets business requirements
Cost	Total expenditure on model inference, including input tokens, output tokens, and cache operations	Controls operational expenses and enables sustainable scaling
Latency	Time elapsed from request initiation to response completion	Impacts user experience and application responsiveness

---

Workshop Structure

This workshop is organized into progressive parts:

Part 1: Basics - Fundamentals

Estimated time: 1.5 hours

Build a solid understanding of tokens, pricing, and optimization strategies.

Topic	Duration	Description
Prompts 101	30 min	Tokens, pricing, TPM/RPM, terminology
Optimization Strategy	45 min	Model selection, prompt design, parameter tuning, basic caching
Langfuse Observability	30 min	LLM tracing, cost tracking, prompt management with Langfuse

Part 2: Developer Journey

Estimated time: 3 hours

Build a production customer support agent while applying progressive optimization techniques.

Topic	Duration	Description
Baseline Agent	20 min	Build unoptimized baseline agent, establish metrics
Quick Wins	20 min	Concise prompts, max_tokens, stop_sequences
Prompt Caching	30 min	System prompt and tool definition caching
LLM Routing	30 min	Route queries to appropriate models by complexity
Guardrails	30 min	Bedrock Guardrails for topic/content filtering
AgentCore Gateway	45 min	Semantic tool search, centralized tool management
Evaluations	30 min	Systematic evaluation across all agent versions

Note: Part 2 requires infrastructure deployment. See 02-developer-journey/README.md for setup instructions.

Part 3: Advanced Concepts

Estimated time: 2-2.5 hours

Advanced prompt engineering techniques and complex caching patterns.

Topic	Duration	Description
Advanced Prompt Engineering	60 min	CoT, Self-Critical, CoD, technique selection etc.
Advanced Prompt Caching	60 min	Multi-checkpoint patterns, cache strategies etc.
TBD	TBD	TBD

---

Prerequisites

Required

• AWS Account with Amazon Bedrock access enabled
• Python 3.10 or higher
• Basic familiarity with Python and Jupyter notebooks

---

Setup Instructions

1. Local Environment Setup

Option A: Using uv (Recommended - Fast!)

Install uv (if not already installed):

macOS/Linux:

# Option 1: Official installer (recommended)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Option 2: Using Homebrew
brew install uv

# Option 3: Using pipx
brew install pipx
pipx install uv

Windows:

powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

Create and activate virtual environment:

# Create virtual environment with Python 3.11
uv venv --python 3.11

# Activate virtual environment
# macOS/Linux:
source .venv/bin/activate
# Windows:
.venv\Scripts\activate

# Install dependencies
uv pip install -r requirements.txt

Option B: Using standard pip

# Create virtual environment
python3 -m venv .venv

# Activate virtual environment
# macOS/Linux:
source .venv/bin/activate
# Windows:
.venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

---

2. Configure AWS Credentials

The workshop supports multiple credential methods:

Method 1: AWS CLI Credentials (Recommended)

If you already have AWS CLI configured, the notebooks will automatically use your credentials:

# No additional setup needed - boto3 uses ~/.aws/credentials

Method 2: Environment Variables (.env file)

If you prefer to use a .env file:

# Copy the example file
cp .env.example .env

# Edit .env and uncomment the AWS credentials
# Then add your actual credentials:
# AWS_ACCESS_KEY_ID=your-actual-access-key-id
# AWS_SECRET_ACCESS_KEY=your-actual-secret-access-key
# AWS_DEFAULT_REGION=us-east-1

The notebooks will automatically load credentials using python-dotenv.

---

3. Test Bedrock Connectivity

python -c "
import boto3
client = boto3.client('bedrock-runtime', region_name='us-east-1')
print('Bedrock connection successful!')
print(f'Region: {client.meta.region_name}')
"

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

---