Skip to content

Day 1 Overview: Foundation & AUTOMAT

Duration: 1.5 hours
Focus: Building prompt engineering foundations and learning your first framework

Learning Objectives

By the end of Day 1, you will:

✅ Understand why prompt engineering matters for R&D efficiency
✅ Apply the AUTOMAT framework to functional scientific tasks
✅ Use conversational learning to build expertise, not just get answers
✅ Protect IP using the Red List Protocol
✅ Practice in the sandbox environment with real scenarios

Session Structure

Part 1: Foundation (30 minutes)

Introduction to Prompt Engineering (15 min) - What is prompt engineering? - Why it matters for materials scientists - Efficiency gains and quality improvements - The learning journey ahead

Sandbox Setup (15 min) - Access your local AI environment - First prompts and experimentation - Understanding model behaviour

Part 2: AUTOMAT Framework (40 minutes)

Framework Introduction (15 min) - Seven components: Audience, User, Task, Output, Method, Assumptions, Tone - Why structured prompts outperform casual requests - Scientific method parallels

Materials Science Applications (15 min) - Literature data extraction - Tensile testing analysis - SEM image interpretation - Complete AUTOMAT examples

Hands-On Practice (10 min) - Build your first AUTOMAT prompt - Test in sandbox - Refine based on output

Part 3: Conversational Learning & Security (20 minutes)

From Transaction to Learning (10 min) - Why "just give me the answer" fails - The power of "why" questions - Building transferable expertise

Responsible AI & IP Protection (10 min) - The Red List Protocol - What never leaves your network - How to sanitise sensitive data - Sandbox vs. external tools

Key Concepts

1. Prompt Engineering is Not About "Talking Nicely to AI"

It's about: - Precision specification (like experimental protocols) - Explicit constraints (like defining experimental boundaries) - Structured frameworks (like the scientific method)

2. AUTOMAT = The Scientific Method for AI

Scientific Method AUTOMAT Framework
Define hypothesis Task
Design experiment Method
Specify measurements Output
Control variables Assumptions
Document rigorously Tone + Audience

3. Conversational Learning > Task Completion

Don't ask: "What's the best solvent for PLA?"
Ask: "Why is DMF preferred? What are the trade-offs? When would alternatives be better?"

Outcome: Transferable decision framework, not just a single answer

4. IP Security is Non-Negotiable

Red List items NEVER go to external AI: - Unpublished data - Patent-pending processes - Customer information - Financial details - Proprietary formulations

Use the sandbox for sensitive work!

What You'll Build Today

Template Library (Started)

By end of Day 1, you'll have:

  • Literature data extraction template
  • Lab notebook formatting template
  • Risk assessment checklist
  • First conversational learning scripts

Real Efficiency Gains

Traditional approach:

  • Literature review: 4-6 hours
  • Protocol formatting: 20-30 min per protocol
  • Multiple iterations to get format right

With AUTOMAT:

  • Literature review: 1-1.5 hours (75% reduction)
  • Protocol formatting: 3-5 min per protocol (85% reduction)
  • First-shot success rate >70%

Today's Exercises

You'll practice with:

  1. Literature Data Extraction (AUTOMAT)

  2. Extract synthesis parameters from 8 papers

  3. Structured output for competitive analysis

  4. Hallucination Hunt

  5. Find 5 deliberate errors in an AI-generated report

  6. Learn verification techniques

  7. Red List Assessment

  8. Evaluate 5 scenarios for IP risk

  9. Practice sanitisation strategies

  10. Prompt Refinement Challenge

  11. Transform a vague prompt into high-quality AUTOMAT prompt

  12. A/B test in sandbox

Materials You'll Use

  • Sandbox environment: http://192.168.1.177:3000
  • Model: Llama 3.2 (3B) - running locally, 100% private
  • Red List Protocol: Reference guide
  • Cheat Sheet: Quick reference

Success Criteria

You're ready for Day 2 when you can:

✅ Explain why structured prompts outperform casual requests
✅ Write a complete AUTOMAT prompt for a functional task
✅ Identify Red List violations in scenarios
✅ Use "why" questions to build learning dialogues
✅ Navigate the sandbox confidently

Common Questions

"Isn't this overkill for simple tasks?"

For one-sentence tasks, yes. But most R&D work isn't simple. Frameworks:

  • Ensure reproducibility
  • Reduce iteration time
  • Create reusable templates
  • Protect against hallucinations

"How long until this feels natural?"

Week 1: Frameworks feel mechanical (checking cheat sheet frequently)
Month 1: Frameworks become intuitive (70% first-shot success)
Month 3: You're building template libraries (5-10 hrs/week saved)

"Can I use ChatGPT/Claude/Gemini for this?"

For public information: Yes (literature, published methods)
For sensitive work: No - use the sandbox

Always check the Red List first!

Pre-Work (If Available)

Recommended reading before the session:

Time: 15-20 minutes

Note: Not required, but provides helpful background

Looking Ahead

Day 2 will build on today's foundation:

  • CO-STAR framework (for strategic communication)
  • Context deep dive (why it's critical)
  • Advanced hallucination detection

Day 3-4 will cover:

  • Technical architecture (how LLMs actually work)
  • Green AI (environmental impact & optimization)
  • Ethics, bias, and responsible deployment

Let's Begin!

Ready to transform how you work with AI?

Next: Introduction to Prompt Engineering →