Exercises | Chapter 2 | DSPy: The Comprehensive Guide

🏋️ Practice Your Signature Skills

These exercises will help you master DSPy signatures. Start with the basics and work up to more complex designs.

💡

Tip: Try to complete each exercise before looking at the solutions. The best way to learn is by doing!

Exercise 1 ⭐ Beginner

Inline Signature Basics

Task: Create inline signatures for the following tasks and test them with sample inputs.

a)

Simple Translation

Create an inline signature that takes text and translates it to Spanish.

b)

Headline Generator

Create an inline signature that takes an article and generates a catchy headline.

c)

Fact Checker

Create an inline signature with two inputs (claim, context) that outputs whether the claim is supported.

Expected Skills:

Using dspy.Predict() with string signatures
Handling single and multiple inputs
Accessing output fields

Exercise 2 ⭐ Beginner

Your First Class-Based Signature

Task: Convert the following inline signature to a class-based signature with proper documentation:

# Convert this inline signature:
qa = dspy.Predict("context, question -> answer")

Requirements:

Create a class named ContextQA
Add a descriptive docstring
Include type hints for all fields
Add descriptions using the desc= parameter
Test with a sample question

Exercise 3 ⭐⭐ Intermediate

Multi-Output Signature

Task: Design a signature for comprehensive email analysis.

Requirements:

Input: email_text (the full email content)
Outputs:
- subject - What the email is about
- sender_intent - What the sender wants (request, inform, complaint, etc.)
- urgency - How urgent (low, medium, high)
- action_required - Boolean indicating if action is needed
- suggested_response - A brief suggested reply

Test your signature with this email:

Hi Team,

The client meeting has been moved to tomorrow at 3 PM instead of Thursday.
Please confirm your availability ASAP as we need to finalize attendees.

This is critical - they're making a decision on our proposal next week.

Thanks,
Sarah

Exercise 4 ⭐⭐ Intermediate

Using Literal Types

Task: Create a content moderation signature using Literal types for constrained outputs.

Requirements:

Input: content - Text to moderate
Outputs:
- category - One of: "safe", "mild", "sensitive", "harmful"
- flags - List of issues detected (violence, hate_speech, adult_content, spam, misinformation)
- confidence - Float between 0.0 and 1.0
- requires_review - Boolean for human review
- explanation - Reasoning for the classification

Test with various content types:

"Check out this amazing recipe for chocolate cake!"
"Breaking: Aliens confirmed to live among us!" (misinformation)
A legitimate news headline of your choice

Exercise 5 ⭐⭐⭐ Advanced

Signature Collection for a Blog Platform

Task: Design a complete set of signatures for a blog platform's AI features.

Create these signatures:

1️⃣

GenerateBlogPost

Takes a topic and key points, generates a complete blog post with title, introduction, body sections, and conclusion.

2️⃣

SEOOptimizer

Takes a blog post and generates SEO metadata: meta title, meta description, keywords, and suggested URL slug.

3️⃣

SocialMediaPosts

Takes a blog post and generates social media content: tweet (280 chars), LinkedIn post, and Instagram caption.

4️⃣

ContentAnalyzer

Takes a blog post and outputs: reading time, difficulty level, target audience, topics covered, and improvement suggestions.

Bonus Challenge: Create a simple script that uses all four signatures in sequence to demonstrate a complete content pipeline.

💡 Exercise Tips

📝

Write Clear Docstrings

The docstring is your main instruction to the LM - make it specific!

🏷️

Use Appropriate Types

Choose types that match your expected output format (str, list, bool, Literal)

📏

Add Constraints in Descriptions

Use descriptions to specify formats, lengths, and valid values

🧪

Test with Edge Cases

Try unusual inputs to see how your signatures handle them

🎉

Great Work!

Completed the exercises? Check your solutions!

View Solutions