Module 3 Assessment - Prompt University

Time Estimate: 60-90 minutes
Passing Score: 80% (20/25 questions + successful project implementation)

Assessment Overview

This assessment evaluates your mastery of advanced prompting techniques covered in Module 3:

Chain of Thought (CoT) prompting
Problem decomposition strategies
Self-refinement and iteration
Ensembling and multi-path reasoning
Tool integration and RAG

This is the culminating assessment for the entire Prompt University course. It combines techniques from all three modules into a comprehensive final project.

Part 1: Multiple Choice Questions (25 questions)

Question 1: Chain of Thought Fundamentals

What is the primary mechanism by which CoT improves reasoning accuracy?

Your Answer
Correct Answer

A) It makes the model run slower, allowing more processing time
B) It breaks down complex problems into explicit intermediate steps
C) It increases the model’s parameter count
D) It accesses external knowledge bases

Question 2: Zero-Shot CoT

Which phrase is most effective for triggering zero-shot Chain of Thought reasoning?

Your Answer
Correct Answer

A) “Think carefully”
B) “Let’s think step-by-step”
C) “Use your knowledge”
D) “Be thorough”

Question 3: When to Use CoT

When is Chain of Thought prompting LEAST beneficial?

Your Answer
Correct Answer

A) Multi-step math problems
B) Simple factual questions
C) Logical reasoning tasks
D) Counter-intuitive problems

Question 4: Least-to-Most Prompting

What is the key principle of least-to-most prompting?

Your Answer
Correct Answer

A) Start with the hardest sub-problem first
B) Solve the simplest sub-problem first and build up
C) Solve all sub-problems simultaneously
D) Skip intermediate steps

Question 5: Problem Decomposition

When should you use problem decomposition?

Your Answer
Correct Answer

A) For all problems, regardless of complexity
B) Only for math problems
C) For hierarchical or multi-domain problems
D) Never, it’s always inefficient

Question 6: Self-Consistency

How does self-consistency improve accuracy?

Your Answer
Correct Answer

A) By generating multiple reasoning paths and using majority vote
B) By making the model more confident
C) By increasing temperature
D) By using longer prompts

Question 7: Iterative Refinement

What is the correct order for the refinement cycle?

Your Answer
Correct Answer

A) Critique → Generate → Refine
B) Generate → Refine → Critique
C) Generate → Critique → Refine
D) Refine → Generate → Critique

Question 8: Self-Critique

What makes a good self-critique?

Your Answer
Correct Answer

A) Vague statements like “could be better”
B) Specific, actionable issues with concrete examples
C) Only positive feedback
D) Focus on style over substance

Question 9: Ensembling Benefits

What is the primary advantage of ensembling multiple approaches?

Your Answer
Correct Answer

A) It’s faster than single approaches
B) Different approaches make different errors, reducing overall error rate
C) It uses less computational resources
D) It’s simpler to implement

Question 10: Voting Mechanisms

When should you use weighted voting instead of simple majority vote?

Your Answer
Correct Answer

A) Always, it’s always better
B) When some methods are more reliable for the specific problem type
C) Never, simple majority is always sufficient
D) Only for math problems

Question 11: RAG Fundamentals

What does RAG stand for and what does it do?

Your Answer
Correct Answer

A) Random Answer Generation - generates random responses
B) Retrieval-Augmented Generation - retrieves relevant info before generating
C) Rapid AI Generation - speeds up response time
D) Recursive Algorithm Generation - creates algorithms recursively

Question 12: RAG Benefits

What is the primary benefit of using RAG?

Your Answer
Correct Answer

A) Faster response times
B) Reduced hallucinations through factual grounding
C) Smaller model size requirements
D) Simpler prompts

Question 13: Knowledge Grounding

What is the most important rule for knowledge-grounded responses?

Your Answer
Correct Answer

A) Always provide an answer, even if uncertain
B) Answer ONLY based on provided context, never infer
C) Make educated guesses when information is missing
D) Prioritize creativity over accuracy

Question 14: Source Citation

Why is source citation important in RAG systems?

Your Answer
Correct Answer

A) It makes responses longer
B) It provides verifiability and traceability
C) It’s required by law
D) It impresses users

Question 15: Tool Integration

What is the main purpose of integrating external tools with LLMs?

Your Answer
Correct Answer

A) To make the system more complex
B) To enable actions and access to real-time/specialized data
C) To slow down response time
D) To increase token usage

Question 16: CoT Accuracy Improvement

By approximately how much can CoT improve accuracy on reasoning tasks?

Your Answer
Correct Answer

A) 5-10%
B) 15-20%
C) 30-50%
D) 70-90%

Question 17: Decomposition vs CoT

How does problem decomposition differ from Chain of Thought?

Your Answer
Correct Answer

A) They’re the same technique
B) Decomposition breaks into sub-problems; CoT shows step-by-step reasoning
C) CoT is always better
D) Decomposition is only for math

Question 18: Refinement Iterations

How many refinement iterations are typically recommended for critical content?

Your Answer
Correct Answer

A) Always just 1
B) 2-3 iterations
C) 10+ iterations
D) Never refine, first draft is best

Question 19: Ensembling Cost

What is the main trade-off when using ensembling?

Your Answer
Correct Answer

A) Reduced accuracy for faster speed
B) Increased computational cost for improved accuracy
C) Simpler implementation for reduced features
D) No trade-offs, it’s always better

Question 20: RAG vs Fine-Tuning

When should you use RAG instead of fine-tuning?

Your Answer
Correct Answer

A) When information changes frequently
B) When you want to modify model behavior permanently
C) When you have unlimited training data
D) Never, fine-tuning is always better

Question 21: Verification in CoT

Why is adding a verification step important in CoT?

Your Answer
Correct Answer

A) It makes the response longer
B) It catches errors in reasoning before finalizing the answer
C) It’s required by the model
D) It impresses users

Question 22: Recursive Decomposition

When is recursive decomposition most useful?

Your Answer
Correct Answer

A) For all problems
B) For self-similar problems where sub-problems resemble the main problem
C) Only for computer science problems
D) Never, it’s too complex

Question 23: Confidence Scoring

What should you do when multiple solutions have low confidence?

Your Answer
Correct Answer

A) Pick one randomly
B) Add more solution methods or request more information
C) Always go with the first solution
D) Give up

Question 24: Tool Error Handling

What’s the best approach when a tool call fails?

Your Answer
Correct Answer

A) Give up immediately
B) Try alternative tools or approaches, explain the limitation to the user
C) Pretend it worked
D) Ignore the error

Question 25: Combining Techniques

Which combination of techniques is most powerful for complex, high-stakes problems?

Your Answer
Correct Answer

A) Just use CoT alone
B) CoT + Decomposition + Self-Refinement + RAG
C) Only use RAG
D) Avoid combining techniques

Part 2: Final Project - Multi-Capability AI Assistant

Capstone Project: This comprehensive project tests your ability to integrate ALL advanced techniques from Module 3.

Project Overview

Build a Multi-Capability AI Assistant that:

Uses Chain of Thought for reasoning
Applies problem decomposition for complex queries
Implements self-refinement for quality assurance
Leverages RAG for factual grounding
Integrates external tools for calculations and data access

System Requirements

Your assistant must handle three types of queries: Type 1: Complex Reasoning (CoT + Decomposition)
Type 2: Knowledge-Intensive (RAG + Source Citation)
Type 3: Action-Requiring (Tool Integration)

Task 1: Complex Reasoning System (30 points)

Build a system that handles multi-step reasoning problems. Requirements:

Use Chain of Thought for step-by-step reasoning
Apply decomposition for problems with 3+ sub-components
Include verification steps
Show confidence levels

System Prompt Template

You are an advanced reasoning assistant that solves complex problems systematically.

Problem: [USER QUERY]

STEP 1: Analyze Complexity
- Is this a single-step or multi-step problem?
- Does it require decomposition?
- What's the core challenge?

STEP 2: Choose Strategy
- Simple problem: Direct CoT
- Complex problem: Decomposition + CoT
- Ambiguous problem: Multiple approaches

STEP 3: Solve with CoT
Let's think step-by-step:
[DETAILED REASONING]

STEP 4: Verify Solution
Check:
- Are all steps logical?
- Does the answer make sense?
- Any calculation errors?

STEP 5: Present Solution
Answer: [FINAL ANSWER]
Confidence: [HIGH/MEDIUM/LOW]
Reasoning: [SUMMARY]

Test Case: “A company wants to reduce costs by 20% while maintaining quality. They spend

500K on materials,

300K on labor, and $200K on overhead. What strategies should they consider?”

Expected Output Structure

STEP 1: Analyze Complexity
This is a multi-faceted business problem requiring:
- Cost analysis across categories
- Strategic recommendations
- Trade-off considerations

STEP 2: Decomposition
Sub-problem 1: Calculate target cost reduction ($200K)
Sub-problem 2: Analyze each cost category
Sub-problem 3: Identify reduction strategies
Sub-problem 4: Assess quality impact

STEP 3: Solve Each Sub-Problem

Sub-problem 1: Target Reduction
- Current total: $1M
- Target reduction: 20% = $200K
- New target: $800K

Sub-problem 2: Category Analysis
- Materials: $500K (50%)
- Labor: $300K (30%)
- Overhead: $200K (20%)

Sub-problem 3: Strategies by Category

Materials ($500K → target $400K):
- Negotiate bulk discounts: -10% = $50K
- Alternative suppliers: -5% = $25K
- Reduce waste: -5% = $25K
Total potential: $100K

Labor ($300K → target $240K):
- Process automation: -10% = $30K
- Efficiency improvements: -5% = $15K
- Flexible scheduling: -5% = $15K
Total potential: $60K

Overhead ($200K → target $160K):
- Renegotiate contracts: -10% = $20K
- Energy efficiency: -5% = $10K
- Space optimization: -5% = $10K
Total potential: $40K

Sub-problem 4: Quality Impact Assessment
- Materials: Medium risk (supplier changes)
- Labor: Low risk (automation improves consistency)
- Overhead: Low risk (no direct quality impact)

STEP 4: Synthesis
Total potential savings: $200K (meets target)
Recommended approach:
1. Start with overhead (lowest risk): $40K
2. Implement labor efficiencies: $60K
3. Negotiate materials carefully: $100K

STEP 5: Verification
✓ Math checks out: $200K = 20% of $1M
✓ Strategies are actionable
✓ Quality risks identified and manageable

Final Answer:
The company can achieve 20% cost reduction through a phased approach:
- Phase 1 (Low risk): Overhead optimization ($40K)
- Phase 2 (Low-medium risk): Labor efficiency ($60K)
- Phase 3 (Medium risk): Materials optimization ($100K)

This strategy minimizes quality risk while achieving the target.

Confidence: HIGH

Task 2: Knowledge-Grounded QA System (30 points)

Build a RAG-based system that answers questions using provided knowledge. Requirements:

Retrieve and use relevant context
Cite sources explicitly
Admit when information is insufficient
Prevent hallucinations

System Prompt Template

You are a knowledge-grounded assistant. Answer questions using ONLY the provided context.

Knowledge Base:
[RETRIEVED DOCUMENTS]

Question: [USER QUERY]

Grounding Rules:
1. Answer ONLY from provided context
2. Cite sources: [Doc X]
3. If information missing: State explicitly
4. Never infer or assume
5. Provide confidence level

Response Format:
Answer: [GROUNDED RESPONSE]

Sources: [LIST OF DOCS USED]

Confidence: [HIGH/MEDIUM/LOW]

Limitations: [IF ANY]

Test Case: Provide 3 documents about a fictional product and answer questions about it.

Sample Implementation

Knowledge Base:

[Doc 1 - Product Specs]:
"The XR-2000 smartwatch features a 1.4-inch AMOLED display, 7-day battery life, 
and water resistance up to 50 meters. It includes heart rate monitoring, GPS tracking, 
and sleep analysis. Price: $299."

[Doc 2 - Customer Reviews]:
"Average rating: 4.5/5 stars from 1,247 reviews. Users praise the battery life 
and display quality. Common complaints include limited app selection and occasional 
GPS accuracy issues."

[Doc 3 - Warranty]:
"1-year manufacturer warranty covers defects. Extended 2-year warranty available 
for $49. Warranty does not cover water damage, physical damage, or battery degradation."

Question: "Is the XR-2000 good for swimming and what's the warranty?"

Response:

Answer: Yes, the XR-2000 is suitable for swimming as it has water resistance up 
to 50 meters [Doc 1]. However, it's important to note that the warranty does NOT 
cover water damage [Doc 3], so while the device is designed for water use, any 
water-related issues would not be covered under warranty.

The watch comes with a 1-year manufacturer warranty that covers defects, with an 
optional 2-year extended warranty available for $49 [Doc 3].

Sources:
- [Doc 1]: Product specifications (water resistance)
- [Doc 3]: Warranty terms

Confidence: HIGH (information directly stated in sources)

Limitations: The documents don't specify what "water resistance up to 50 meters" 
means in practical terms (e.g., suitable for swimming vs. diving), though 50 meters 
typically indicates swimming is safe.

Task 3: Tool-Integrated Assistant (40 points)

Build a system that uses external tools to accomplish tasks. Requirements:

Identify when tools are needed
Execute tools with proper parameters
Handle tool errors gracefully
Combine tool outputs with reasoning

System Prompt Template

You are an assistant with access to external tools.

Available Tools:
1. calculator(expression) - Evaluates math expressions
2. search(query) - Searches knowledge base
3. get_data(source, parameters) - Retrieves data from sources

User Request: [QUERY]

Process:

STEP 1: Analyze Request
What information/actions are needed?
Which tools are required?

STEP 2: Plan Tool Usage
Tool sequence:
1. [TOOL] with [PARAMETERS]
2. [TOOL] with [PARAMETERS]

STEP 3: Execute Tools
Tool: [NAME]
Input: [PARAMETERS]
Output: [RESULT]

[Repeat for each tool]

STEP 4: Synthesize Response
Using tool outputs:
[COMPREHENSIVE ANSWER]

STEP 5: Verify
✓ All tools executed successfully?
✓ Results make sense?
✓ Answer is complete?

Final Response: [ANSWER]

Test Case: “Calculate the compound interest on $10,000 invested at 6% annually for 5 years, then compare it to simple interest for the same period.”

Expected Output

STEP 1: Analyze Request
Need to:
- Calculate compound interest
- Calculate simple interest
- Compare the two

Tools needed: calculator

STEP 2: Plan Tool Usage
1. Calculate compound interest: A = P(1 + r)^t
2. Calculate simple interest: I = Prt
3. Calculate difference

STEP 3: Execute Tools

Tool: calculator
Input: "10000 * (1 + 0.06)^5"
Output: 13382.26

Tool: calculator
Input: "10000 * 0.06 * 5"
Output: 3000.00

Tool: calculator
Input: "13382.26 - (10000 + 3000)"
Output: 382.26

STEP 4: Synthesize Response

Compound Interest:
- Formula: A = P(1 + r)^t
- Calculation: $10,000 × (1.06)^5 = $13,382.26
- Interest earned: $3,382.26

Simple Interest:
- Formula: I = Prt
- Calculation: $10,000 × 0.06 × 5 = $3,000
- Total amount: $13,000

Comparison:
- Compound interest earns $382.26 more than simple interest
- This represents a 12.7% higher return
- The difference comes from "interest on interest" in compound calculation

STEP 5: Verify
✓ Compound interest formula correct
✓ Simple interest formula correct
✓ Comparison calculation accurate

Final Response:
Over 5 years, compound interest ($3,382.26) earns $382.26 more than simple 
interest ($3,000) on a $10,000 investment at 6% annual rate. Your final amount 
would be $13,382.26 with compound interest versus $13,000 with simple interest.

Scoring Rubric

Multiple Choice (50 points)

2 points per question
20/25 correct required to pass (40/50 points)

Final Project (50 points)

Task 1: Complex Reasoning (15 points)

Proper use of CoT: 5 points
Effective decomposition: 5 points
Verification and confidence: 5 points

Task 2: Knowledge-Grounded QA (15 points)

Strict grounding (no hallucinations): 5 points
Proper source citation: 5 points
Handling missing information: 5 points

Task 3: Tool Integration (20 points)

Correct tool identification: 5 points
Proper tool execution: 5 points
Error handling: 5 points
Result synthesis: 5 points

Total: 100 points
Passing Score: 80 points

Evaluation Criteria

Technical Accuracy

Correct application of techniques and accurate results

Integration

Effective combination of multiple techniques

Robustness

Handles edge cases and errors gracefully

Clarity

Clear reasoning and well-structured outputs

Course Completion

Complete all 25 multiple-choice questions

Implement Task 1: Complex Reasoning System

Implement Task 2: Knowledge-Grounded QA

Implement Task 3: Tool-Integrated Assistant

Test your system with provided test cases

Verify all components work together

Congratulations!

Upon completing this assessment, you will have demonstrated mastery of:

✅ Foundational prompting principles (Module 1)
✅ Task-specific prompting patterns (Module 2)
✅ Advanced prompting techniques (Module 3)

You’re now equipped to build sophisticated AI systems using state-of-the-art prompting techniques!

View Course Certificate

Claim your Prompt University completion certificate

Module 1

Module 2

Module 3

​Assessment Overview

​Part 1: Multiple Choice Questions (25 questions)

​Question 1: Chain of Thought Fundamentals

​Question 2: Zero-Shot CoT

​Question 3: When to Use CoT

​Question 4: Least-to-Most Prompting

​Question 5: Problem Decomposition

​Question 6: Self-Consistency

​Question 7: Iterative Refinement

​Question 8: Self-Critique

​Question 9: Ensembling Benefits

​Question 10: Voting Mechanisms

​Question 11: RAG Fundamentals

​Question 12: RAG Benefits

​Question 13: Knowledge Grounding

​Question 14: Source Citation

​Question 15: Tool Integration

​Question 16: CoT Accuracy Improvement

​Question 17: Decomposition vs CoT

​Question 18: Refinement Iterations

​Question 19: Ensembling Cost

​Question 20: RAG vs Fine-Tuning

​Question 21: Verification in CoT

​Question 22: Recursive Decomposition

​Question 23: Confidence Scoring

​Question 24: Tool Error Handling

​Question 25: Combining Techniques

​Part 2: Final Project - Multi-Capability AI Assistant

​Project Overview

​System Requirements

​Task 1: Complex Reasoning System (30 points)

​Task 2: Knowledge-Grounded QA System (30 points)

​Task 3: Tool-Integrated Assistant (40 points)

​Scoring Rubric

​Multiple Choice (50 points)

​Final Project (50 points)

​Evaluation Criteria

Technical Accuracy

Integration

Robustness

Clarity

​Course Completion

​Congratulations!

View Course Certificate

Assessment Overview

Part 1: Multiple Choice Questions (25 questions)

Question 1: Chain of Thought Fundamentals

Question 2: Zero-Shot CoT

Question 3: When to Use CoT

Question 4: Least-to-Most Prompting

Question 5: Problem Decomposition

Question 6: Self-Consistency

Question 7: Iterative Refinement

Question 8: Self-Critique

Question 9: Ensembling Benefits

Question 10: Voting Mechanisms

Question 11: RAG Fundamentals

Question 12: RAG Benefits

Question 13: Knowledge Grounding

Question 14: Source Citation

Question 15: Tool Integration

Question 16: CoT Accuracy Improvement

Question 17: Decomposition vs CoT

Question 18: Refinement Iterations

Question 19: Ensembling Cost

Question 20: RAG vs Fine-Tuning

Question 21: Verification in CoT

Question 22: Recursive Decomposition

Question 23: Confidence Scoring

Question 24: Tool Error Handling

Question 25: Combining Techniques

Part 2: Final Project - Multi-Capability AI Assistant

Project Overview

System Requirements

Task 1: Complex Reasoning System (30 points)

Task 2: Knowledge-Grounded QA System (30 points)

Task 3: Tool-Integrated Assistant (40 points)

Scoring Rubric

Multiple Choice (50 points)

Final Project (50 points)

Evaluation Criteria

Course Completion

Congratulations!