Skip to main content

​
Best Practices

​
Writing Effective Prompts

✅ Good: “Design per ACI 318-19, Chapter 17”❌ Avoid: “Design per ACI code”Different code editions have different requirements. Always specify the exact version.
✅ Good: Include all necessary parameters (dimensions, materials, loads, code requirements)❌ Avoid: Vague descriptions like “standard office building loads”The more complete your inputs, the more accurate and useful the Agent’s output.
✅ Good: “First, find the relevant code sections. Then, plan the calculation approach. Finally, generate the Mathcad sheet.”❌ Avoid: “Do everything for this entire building”Complex projects benefit from a step-by-step approach with review points.
✅ Good: “Generate a Mathcad sheet with an input block at the top, calculations in the middle, and a summary table at the bottom”❌ Avoid: “Give me calculations”Tell the Agent how you want the deliverable structured.
✅ Good: “Use our firm’s ‘Standard_Column_Design.mcdx’ template from the private knowledge base”❌ Avoid: “Use our standard template” (if you haven’t uploaded it)Explicitly reference documents in your private knowledge base by filename.

​
Workflow Tips

1

Start with Code Research

Before generating calculations, ask the Agent to identify relevant code sections and requirements. This helps you verify the approach.Example: “What ACI 318-19 sections apply to two-way slab design with drop panels?”
2

Review Intermediate Steps

For complex calculations, review the Agent’s planning and approach before asking it to generate the full Mathcad sheet.Example: “Before creating the Mathcad sheet, show me the calculation outline and approach you’ll use.”
3

Iterate and Refine

Don’t expect perfection on the first try. Use follow-up prompts to refine calculations, add checks, or adjust formatting.Example: “Can you add a deflection check to the existing calculation? Use the service loads from the input block.”
4

Leverage Templates

Create and reuse templates for common calculation types. This maintains consistency and speeds up future work.Example: “Using the ‘Column_Template.mcdx’ from my knowledge base, create a design for a 16x16 column with the following loads…”
5

Perform Engineering Review

Always review Agent-generated calculations with the same rigor you would apply to any engineering deliverable. Check assumptions, verify code references, and validate results.

​
Common Patterns

​
Pattern 1: Research → Plan → Execute

Prompt: "What are the ACI 318-19 requirements for designing
a transfer beam? List all relevant sections."

​
Pattern 2: Template → Customize → Validate

Prompt: "Using our 'Footing_Design_Template.mcdx' from the
private knowledge base, create a new calculation for a
square footing."

​
Pattern 3: Migrate → Enhance → Standardize

Prompt: "Convert the attached hand calculation (PDF) to a
clean Mathcad Prime 9 file. Preserve all original steps."

​
Tips for Success

Attach Reference Files

Attach relevant PDFs, images, or existing Mathcad files to your prompts. The Agent can read and reference them in its work.

Use Follow-Up Prompts

Don’t try to get everything perfect in one prompt. Use follow-ups to refine and adjust the output.

Build a Template Library

Create standard templates for common calculations and save them to your Private Knowledge Base for reuse.

Provide Context

Mention project context when relevant (high seismic, wind-controlled, etc.) to help the Agent prioritize the right checks.

Review Before Each Step

For multi-step workflows, review intermediate outputs before proceeding to ensure you’re on the right track.

Report Issues

If you find errors or unexpected behavior, report them to bhosh@stru.ai with details. This helps improve the Agent for everyone.

​
What to Avoid

Don’t skip engineering review: Never submit Agent-generated calculations without thorough professional review by a licensed engineer.
Don’t provide incomplete inputs: Vague or incomplete information leads to assumptions that may not match your project requirements.
Don’t ignore warnings: If the Agent warns about assumptions, limitations, or areas needing review, address them before using the deliverable.
Don’t expect mind-reading: The Agent can’t know your project specifics unless you provide them. Be explicit about requirements, constraints, and preferences.

​
Getting Better Results

The more you use Stru AI, the better it understands your preferences through its self-learning architecture. Here are ways to accelerate this:
  1. Be consistent with terminology and formatting preferences
  2. Provide feedback when the output doesn’t match your expectations
  3. Save and reuse successful prompts for similar tasks
  4. Build templates that capture your firm’s standards
  5. Use the Private Knowledge Base to give the Agent access to your firm’s standards
Remember: The Agent learns your style over time. Early projects may require more refinement, but subsequent projects will align better with your preferences automatically.