Skip to content

Introduction to Prompt Engineering

Welcome to Day 1!

Before we dive into frameworks, let's establish why you're here and what prompt engineering actually means for materials scientists.

What is Prompt Engineering?

Simple definition: The practice of designing inputs (prompts) to get specific, high-quality outputs from AI models.

For materials scientists: The skill of transforming vague requests into precise instructions that yield scientifically rigorous results.

The Problem We're Solving

Scenario: Literature Review

You need: Synthesis parameters from 20 papers on PLA/graphene composites

Approach 1: Casual AI Use 

User: "Summarise these papers on PLA"
AI: [Generic summary of topics, no specific data]
User: "I need synthesis parameters"
AI: [Mentions some parameters, inconsistent format]
User: "Can you put that in a table?"
AI: [Creates table, missing key columns]
User: "Add temperature and pressure columns"
AI: [Adds columns, but mixes units]
User: "Use consistent units"
... [5 more iterations]

Time: 20 minutes
Quality: Mediocre
Reproducible: No

Approach 2: Prompt Engineering 

Act as Materials Science Data Analyst.

Task: Extract synthesis parameters from attached 20 papers on 
PLA/graphene nanocomposites.

Output: Markdown table with columns:

- Paper (Author Year)
- Synthesis Method
- Temperature (°C)
- Pressure (kPa)
- Graphene Loading (wt%)
- Yield (%)

Constraints:

- If parameter not reported, mark "NR"
- Only include electrospinning methods
- Convert all temperatures to Celsius

Tone: Technical, data-focused

Time: 3 minutes
Quality: High-fidelity
Reproducible: Yes (template for future use)

Why This Matters for R&D

Efficiency Gains

Before prompt engineering:

  • Literature review: 8 hours → 2 hours (75% reduction)
  • Protocol formatting: 30 min/protocol → 5 min (83% reduction)
  • Experimental design brainstorming: 2 hours → 30 min (75% reduction)

After prompt engineering:

  • Same quality or better
  • Reproducible workflows
  • Template libraries for recurring tasks

Quality Improvements

Structured prompts lead to:

  • ✅ Consistent output format
  • ✅ Fewer hallucinations (explicit constraints)
  • ✅ Verifiable results (cite sources, show calculations)
  • ✅ Suitable for professional use

IP Protection

Prompt engineering includes security:

  • Know what data to never share (Red List)
  • Sanitise sensitive information before processing
  • Use local models for confidential work
  • Verify all outputs before use

The Learning Journey

Week 1: Mechanical Application

You'll use frameworks by reference, checking the cheat sheet for each component.

Normal: Prompts feel structured but mechanical.

Month 1: Natural Integration

Frameworks become intuitive. You'll think in terms of audience, constraints, and output format automatically.

Achievement: First-shot success rate >70%

Month 3: Template Building

You'll have created 10-20 templates for your common tasks. New colleagues will ask for your prompts.

Achievement: Time savings 5-10 hours/week

Month 6: Teaching Others

You'll identify new applications, teach frameworks to colleagues, and innovate processes.

Achievement: Team-wide efficiency gains

Today's Focus: AUTOMAT Framework

The AUTOMAT framework gives you a systematic method for crafting scientific prompts.

Seven components that ensure complete specification:

  1. Audience
  2. User Persona
  3. Task
  4. Output
  5. Method
  6. Assumptions
  7. Tone

Think of it as: The scientific method for AI prompting.

What Makes a Good Prompt?

Poor Prompt Characteristics

❌ Vague task: "Help me with this data"
❌ No output specification: AI guesses format
❌ Missing constraints: AI fills gaps with guesses
❌ Generic context: AI provides generic answers
❌ No verification guidance: Hallucinations go undetected

Good Prompt Characteristics

✅ Specific task: "Extract tensile strength values from these 7 datasets"
✅ Defined output: "Table with columns: Sample ID, Strength (MPa), Modulus (GPa)"
✅ Explicit constraints: "If data missing, mark 'NR'—do not estimate"
✅ Relevant context: "For QC documentation, ISO 527 standards"
✅ Verification built-in: "Flag any values >2 SD from mean"

Common Misconceptions

Misconception 1: "AI is smart enough to figure it out"

Reality: AI is a pattern-matching prediction engine. Ambiguity leads to guessing, which leads to inconsistent or wrong outputs.

Solution: Be explicit. Over-specify rather than under-specify.

Misconception 2: "Longer prompts are always better"

Reality: Relevant detail is valuable. Irrelevant context is noise.

Solution: Include task-relevant information only. Context should help, not distract.

Misconception 3: "One framework for everything"

Reality: Different frameworks suit different tasks.

Solution: AUTOMAT for functional tasks, CO-STAR for strategic communication (you'll learn CO-STAR tomorrow).

Misconception 4: "Prompt engineering replaces expertise"

Reality: Prompt engineering amplifies your expertise. You still need domain knowledge to:

  • Identify what questions to ask
  • Verify outputs for accuracy
  • Spot hallucinations
  • Apply results appropriately

Solution: Use AI to augment, not replace, your scientific judgment.

The Materials Science Advantage

You already think like a prompt engineer!

Scientific method:

  1. Define hypothesis (Task)
  2. Design experiment (Method)
  3. Specify measurements (Output)
  4. Control variables (Constraints)
  5. Document rigorously (Format)

Prompt engineering:

  1. Define task
  2. Specify method
  3. Define output format
  4. Set constraints
  5. Structure precisely

Parallel thinking → Faster learning curve for scientists!

Your Toolkit After Today

By end of Day 1, you'll have:

📘 AUTOMAT Framework - Systematic method for functional tasks
🧠 Conversational Learning - Build expertise, don't just get answers
🚫 Red List Protocol - Protect IP whilst using AI
💻 Sandbox Access - Practice with sensitive data safely
⚡ Templates Started - Foundation for your prompt library

Let's Begin

Now that you understand WHY prompt engineering matters and WHAT it can do for you...

Let's learn HOW to do it.

Next: The AUTOMAT Framework: Your systematic method for scientific prompts →