BrainBank
AI Classroom/Best PracticesAgentic AI

Aug15 LLM CHEATSHEET-Role-Dimension-Graph

8/15/2026, 10:56:19 PM · updated 8/15/2026, 10:57:45 PM

#best-practices

A visual cheatsheet mapping how to structure LLM prompts using a Role-Dimension-Graph approach, connecting system roles, prompt dimensions (e.g., format, tone, constraints), and output graphs for reliable prompt design.

CHEATSHEET-Role-Dimension-Graph.html

Learning map

Learning Map: Role-Dimension-Graph for Prompt Design

Stage 1 — Foundations

  • Understand the concept of role in prompting (who the LLM should act as)
  • Learn what dimensions are (parameters that shape behavior: format, tone, constraints, audience)
  • Grasp how a graph visualizes the relationships between roles and dimensions

Stage 2 — Mapping Roles

  • Identify common LLM roles (e.g., expert consultant, tutor, reviewer, creator)
  • Practice assigning explicit system prompts that define role
  • Learn to chain roles for multi-step workflows

Stage 3 — Defining Dimensions

  • Enumerate key dimensions: output format, style depth, scope breadth, safety guardrails
  • Combine multiple dimensions into structured prompt templates
  • Discover which dimension combinations produce the most reliable outputs

Stage 4 — Building Graphs

  • Translate role + dimension combos into a visual or tabular graph
  • Use the graph to audit and debug existing prompts
  • Iterate: replace broken linkages in the graph with working ones

Get hands-on — step by step

  1. Create a blank grid labeled 'Role' on one axis and 'Dimension' on the other.
  2. Fill the role axis with at least three roles you commonly need (e.g., 'Technical Writer', 'Code Reviewer', 'Educator').
  3. Fill the dimension axis with four parameters: format, tone, scope, and constraints.
  4. For each cell in the grid, write a one-line system prompt that pairs a role with a dimension (e.g., as a Code Reviewer, keep your response under 200 words and use bullet points).
  5. Test each combination by running a real prompt against an LLM and note whether the output matches expectation.
  6. Connect cells in the grid where multiple dimensions produce similar outputs — these are redundant links you can merge or discard.
  7. Build a final master chart that highlights your working combinations as green nodes and weak or broken ones as red nodes, then use it like a lookup table whenever you draft a new prompt.

Top 3 sources

  1. 1
    Prompt Engineering Guide by Prompt Engineering Org

    Comprehensive, freely available guide covering prompt patterns, role-based design, and dimension parameters for structuring reliable LLM interactions.

    https://www.promptingguide.ai/

  2. 2
    Google's 'Build a Large Language Model (From Scratch)'

    Hands-on book and codebase that explains how to design, train, and evaluate prompts — includes role-based instruction templates and structured output techniques.

    https://github.com/d2t1/llm-from-scratch

  3. 3
    Andrew Ng's ChatGPT Prompt Engineering for Developers (DeepLearning.AI)

    Official DeepLearning.AI mini-course with practical prompt patterns, including role assignment and dimension-based constraint techniques taught by LLM experts.

    https://www.deeplearning.ai/short-courses/chatgpt-prompt-engineering-for-developers/

Links are AI-suggested — worth a quick sanity check before diving in.