Project Management Training

Prompt Engineering for Australian Construction Management

A Starter Pack for AI Generated Prompting

Introduction

The goal is to enhance the quality and relevance of responses from generative AI models for tasks in Australian construction management. Applying these best practices can significantly improve AI-generated content.

Be Concise

Not Recommended (unnecessarily verbose):

“Considering the current climate and the challenges faced by the Australian construction industry, what do you believe would be an excellent name for a new company specializing in sustainable, eco-friendly residential builds in regional Victoria, specifically focusing on bushfire-prone areas?”

Recommended (concise):

“Suggest a name for a sustainable residential construction company in regional Victoria specializing in bushfire-resistant homes.”

Be Specific and Well-Defined

Not Recommended (generic):

“Tell me about Australian building standards.”

Recommended (specific and well-defined):

“Generate a list of key compliance requirements from the National Construction Code (NCC) related to fire safety for multi-story residential buildings in Queensland.”

Ask One Task at a Time

Not Recommended (multiple tasks):

“What are the common challenges in managing remote construction sites in the Australian outback and what are the primary environmental regulations for earthworks in Western Australia?”

Recommended (single tasks):

“What are the common challenges in managing remote construction sites in the Australian outback?”

“What are the primary environmental regulations for earthworks in Western Australia?”

Turn Generative Tasks into Classification Tasks

This reduces output variability.

Generative task (higher variability):

“Recommend a suitable material for the exterior cladding of a new school building in a cyclone-prone area of Northern Territory, considering durability and cost.”

Classification task (reduced variability):

“For the exterior cladding of a new school building in a cyclone-prone area of Northern Territory, which of these materials is most suitable, and why: a) Brick veneer, b) Fibre cement sheeting, c) Lightweight insulated panels?”

Improve Response Quality with Examples (Shots)

Zero-shot prompting

Offers more creativity.

“Identify the type of construction project based on this description: ‘Development of 50 new homes and associated infrastructure.'”

One-shot prompting

Leads to more predictable and consistent answers.

“Identify the type of construction project based on the description:Project: ‘Construction of a new hospital wing with specialized surgical suites.’Type: Healthcare construction.Project:”

Few-shot prompting

Also provides more predictable and consistent answers.

“Identify the type of construction project based on the description:Project: ‘Construction of a new hospital wing with specialized surgical suites.’Type: Healthcare construction.Project: ‘Upgrade of a wastewater treatment plant to meet new environmental standards.’Type: Infrastructure construction.”

Caution: Watch Out for Hallucinations

Generative AI models can sometimes produce confident but incorrect information due to limited memorization and a lack of real-time data access.

Example: Asking for the exact current market price of building materials like steel rebar in Sydney may result in a fabricated price. Always verify critical information.

Use System Instructions to Guardrail the Model

You can give a large language model system instructions to define its role and minimize irrelevant or fabricated responses.

Example for an Australian construction chatbot:

Relevant Query:

“What are the standard project phases for a large-scale commercial build in Melbourne, and what permits are typically required?”

Irrelevant Query:

“What is the best type of Australian native plant for a home garden?”