INITIALIZING BUNKROS IDENTITY LAB
LOC UNDERGROUND
SYS --:--:--
🎓 14-Module Masterclass

Prompt Engineering
Masterclass

This isn't a list of templates. It's a mental operating system for working with AI — teaching you to think, structure, and reason from first principles.

Why most prompts fail

They treat prompting as clever wording. Real prompt engineering is about intent design, constraint systems, and cognitive scaffolding.

Intent Constraints Scaffolding Feedback Psychology
Begin Your Journey
Scroll to explore
Module 01

Mental Models of AI

Before you can prompt effectively, you must understand how language models actually "think" — and why that's fundamentally different from human cognition.

Token Prediction

LLMs don't understand — they predict. Each word is a probability choice based on all previous context.

Interactive: Click tokens to see prediction flow

The capital of France is Paris
Context Window 4K tokens
⚠️ Key Insight

The model has no persistent memory between conversations. Everything it "knows" must be in the current context.

Module 02

Anatomy of a Prompt

Every effective prompt contains five core components. Understanding them lets you diagnose failures and construct solutions.

Deconstructed Prompt
ROLE You are an expert data analyst...
TASK Analyze the following sales data...
CONTEXT We're a B2B SaaS company in Q4...
CONSTRAINT Keep under 200 words, focus on trends...
FORMAT Return as bullet points with percentages...

Toggle components to see effect:

Prompt Strength: Medium

Missing constraints and format specifications may lead to verbose, unstructured output.

Module 03

Instructions vs Context vs Data

The most common prompt failure: mixing these three layers. Learn to separate them for predictable results.

❌ Mixed

"Summarize this email about the Q3 budget meeting where Sarah mentioned we need to cut costs by 15%"

✓ Separated

"TASK: Summarize
CONTEXT: Q3 budget meeting
DATA: [email content]"

1
Instructions

What you want the model to DO

2
Context

Background the model needs to KNOW

3
Data

Raw material to PROCESS

Module 04

Failure Modes & Debugging

When prompts fail, they fail in predictable ways. Learn to recognize patterns and fix them systematically.

Toggle errors to inject into prompt:

Test Prompt You are a helpful assistant. Analyze the data and provide insights about customer behavior patterns. Return structured output.
🟢 Baseline: Functional

This prompt should work, but lacks specificity. Outputs may vary significantly between runs.

Module 05

Prompt Patterns

Four foundational patterns form the basis of almost all effective prompts. Know when to use each.

Zero-Shot Classify the sentiment of this review as positive, negative, or neutral: "The product exceeded my expectations!"
When to Use

Best for simple, well-defined tasks where the model's training covers the domain. Fast but may lack nuance.

Strengths

Fast, simple, low token cost

Weaknesses

Less control, variable output

Module 06

Multi-Step Systems

Complex tasks require multiple prompts working together. Design systems, not single prompts.

1
Extract

Pull key information from raw input

2
Analyze

Process extracted data with logic

3
Synthesize

Combine insights into output

4
Validate

Check output against criteria

💡 Key Principle

Each step should have a clear input and output. The model shouldn't need to remember previous steps — you pass the result forward explicitly.

Module 07

Constraints & Control

The difference between amateur and professional prompts: precise control over output characteristics.

Length Constraint ~150 words
Formality Level Professional
Creativity Balanced
Generated Constraint Block Respond in approximately 150 words. Use professional language suitable for business communication. Balance factual accuracy with engaging presentation.
Hard Constraints

Exact word counts, specific formats, required sections

Soft Constraints

Tone suggestions, style preferences, general guidance

Module 08

Model-Specific Prompting

The same prompt behaves differently across models. Learn each model's "personality" for optimal results.

GPT-4
Precise, follows instructions
Claude
Nuanced, conversational
Gemini
Multimodal, creative
Llama
Open, customizable
Optimized for GPT-4 You are an expert analyst. Follow these instructions precisely: 1. Analyze the data 2. Return exactly 3 bullet points 3. Each bullet: max 20 words Do not include any preamble or conclusion.
💡 GPT-4 Tip

GPT-4 excels at following explicit, numbered instructions. Be direct and specific. Avoid conversational language in the prompt.

Module 09

Prompting for Work

Professional contexts require reliability, not creativity. Build prompts that produce consistent, defensible output.

Email Template Draft a professional email to [RECIPIENT] regarding [TOPIC]. Tone: Courteous but direct Length: 3-4 paragraphs Include: Clear call to action Avoid: Jargon, passive voice Context: [BACKGROUND]
⚠️ Guardrail

Never let AI make final decisions on critical matters. Use it for drafts, analysis, and options — but humans verify and approve.

Module 10

Prompting for Creativity

Getting original output from AI requires strategic constraint. Paradoxically, more boundaries often yield more creativity.

Generic Prompt

"Write a creative story about space"

Constrained Prompt

"Write 200 words about a janitor on a space station who discovers a hidden message. Use only dialogue. Set during their lunch break."

🎨 Creativity Levers
Perspective shift Format constraint Unusual combo Word limits Specific details
Originality Prompt Avoid: [list 5 common tropes in this genre]
Instead: Find an unexpected angle by combining [unexpected element A] with [unexpected element B].
Constraint: The protagonist cannot be [typical protagonist type].
Module 11

Ethics & Responsibility

With power comes responsibility. Every prompt you write has potential consequences beyond the immediate output.

A colleague asks you to prompt AI to write performance reviews. What's the most responsible approach?
Generate complete reviews and submit as-is
Use AI for structure, human writes content
AI drafts, human heavily edits and verifies
Refuse — reviews should be fully human
!
Bias Amplification

AI inherits and can amplify training biases

!
Dependency Risk

Over-reliance can erode human judgment

!
Attribution

When should AI assistance be disclosed?

Module 12

Prompt Library

Not templates to copy — examples to learn from. Each prompt demonstrates principles from this masterclass.

Decision Analysis Framework Work

Multi-step prompt for evaluating business decisions with pros, cons, and risk assessment.

Originality Breaker Creative

Constraint system that forces novel combinations and avoids common tropes.

Data Insight Extractor Analysis

Structured prompt for pulling patterns and anomalies from datasets.

Meeting Synthesizer Work

Transform meeting notes into actionable summaries with clear ownership.

🎓
Module 13

Mastery Achieved

You've completed the foundational journey. You now think about prompts as systems, not sentences.

🧠 Your New Mental Model
  • Understand how LLMs predict, not understand
  • Separate instructions, context, and data
  • Debug prompts systematically
  • Apply patterns to any task
  • Consider ethics in every prompt
Review from Start