ProjectHelm
Home / Documentation / Prompting & Working with AI
AI Assistant

Prompting & Working with AI

Best practices for instructing the ProjectHelm AI assistant to generate precise schemas, validation expressions, and automations.

Updated September 2026 • 4 min read • .NET 10 & C#

Think in Entities & Relations

ProjectHelm is not an unstructured LLM script writer; it is a declarative schema compiler. You get the best results when you describe your problem in terms of real-world concepts:

❌ Vague Prompt

"Make me a good CRM for my sales team with nice buttons and modern design."

✅ Schema-Driven Prompt

"Create a CRM with Lead (Name, Email, Value as Money, Status), Contact, and Deal entities. Deals belong to Leads. When Deal reaches ClosedWon, mark Lead as Customer."

Specifying Business Rules & Locks

You can specify data integrity rules directly in your request:

  • Validation Rules: "Ensure EndDate is always after StartDate" or "DiscountPercent cannot exceed 25%".
  • Role-Based Locks: "Only users with the 'Manager' role can edit the UnitPrice field".
  • State Locks: "Once an Invoice stage becomes 'Paid', lock the entire record against edits".