Prompt Engineering for New Product Development with LLMs

Picture genrated by ChatGPT

New product development is crucial for driving revenue and growth in any business. It introduces innovative solutions to the market and benefits society by enhancing quality of life and productivity.

The process of new product development begins with a well-defined product strategy, followed by ideation, and culminates in a successful market launch.

The landscape of new product development has transformed dramatically over the past few decades, driven by the emergence of new execution frameworks, the rise of digitalization, the adoption of predictive machine learning models, and the latest advancements in large language models (LLMs).

A Large Language Model (LLM) is a part of artificial intelligence system that can understand and generate human-like text. It’s trained on vast amounts of data, allowing it to answer questions, create content, and assist with tasks by predicting and producing relevant language based on the input it receives. Table-1 has summarized the various applications of LLM on new product development.

Ideation	Concept design	Detail design	Verification	Validation	Limited production
•    Idea generation	•    Technical specification	•    Tech specs for sub system and component	•    testing automation	•    User validation	•    Workflow automation
•    Business case	•    Generating requirement s	•    Design optimization	•    Virtual prototype	•    Service validation
•    Technical feasibility analysis	•    Collaboration and communication	•    Automatic documentation creation	•    Digital twin	•    Performance evaluation
•    Market research	•    Concept scoping and selection	•    Software, coding, automation		•    Test cases generation based on risk
•    Risk management	Design for excellence	•  Design process automation

Table-1 Application of LLM for new product development

The bigger question mark “How can we interact with LLMs to get desire results?”. LLMs rely on well-crafted inputs to deliver accurate and effective results. This process, known as prompt engineering, involves designing these inputs with the right combination of simple text and domain expertise. By doing so, prompt engineering plays a crucial role in new product development, helping to unlock the full potential of LLMs to create more innovative and superior products.

There are many books and references available for prompt engineering. Here is a framework with best practices  of using prompt engineering with some case studies from digital X-ray system.

Framework for Prompt Engineering

Figure -1 Basic framework of  prompt development (PD) for NPD

a. Problem Definition

Start by clearly defining the problem you want the LLM to address. This involves identifying the objective, the type of output needed, and any specific constraints.

Example:

• Problem: "How can we reduce radiation exposure in a Digital X-ray system?"

• Output Required: "Potential solutions that are cost-effective and comply with regulatory standards."

b. Crafting the Initial Prompt

The next step is to frame the input that will be provided to the model. Start with a basic prompt and gradually refine it for better results.

Basic Prompt:

• "Suggest ways to reduce radiation exposure in a Digital X-ray system."

Refined Prompt:

• "As a medical device engineer, suggest three cost-effective ways to reduce radiation exposure in a Digital X-ray system while maintaining image quality, and ensuring the design complies with FDA regulations."

c. Iterative Refinement

Prompt engineering often involves iterative refinement. You try a prompt, evaluate the response, and refine the input based on the feedback. The key is to keep narrowing the focus or expanding where necessary.

Example Refinement:

• "Can you provide more details on the materials or components that could be used to optimize radiation shielding in X-ray systems?"

d. Setting the Role and Context

Defining the role of the LLM helps in receiving more contextual and relevant responses. This might include asking the model to act as an expert in a specific domain.

Example Role Prompt:

• "As a regulatory expert, explain the key compliance challenges to be considered when implementing these radiation reduction techniques in a new X-ray system."

e. Applying Constraints

Applying specific constraints ensures that the model’s response fits within practical limits, such as budget, time, or other resources.

Example:

• "Suggest solutions under a $10,000 budget for upgrading a Digital X-ray system to reduce radiation exposure in paediatric care."

This framework can help to crafted the prompts as per problem statement. We have outlined a best practice to be used for refining the prompt engineering for LLM tools like ChatGPT.

Best practice for the prompt engineering

1.     Iterative testing and refining - One of the most effective ways to improve prompt engineering is through an iterative process of testing and refining.

2.     Balancing user intent and model creativity- While ChatGPT can generate creative responses, it is crucial to balance user intent with model creativity.

3.     Harness external resources and API - In some cases, ChatGPT may not have sufficient knowledge or accuracy to respond to user queries effectively. To address this, prompt engineering can be augmented by harnessing external APIs

4.     Ensuring ethical uses and biases - Many complex product developments follows strict regulation and regulatory norms. Safety is the most critical concern for many products including aerospace and healthcare. An efficient use of prompts by avoiding biases and keeping regulation norms intact may be good way to explore concepts.

5.     Hallucination -  An AI hallucination happens when the large language model (LLM) behind an AI system creates false or misleading information. This can often cause people to make wrong decisions based on incorrect details. Domain expertise is necessary to understand the hallucination. One must validate and deep dive the ideas and analysis provided by LLM to minimise the hallucination.

6.     Learning /example to generate better response-

·      Few-shot learning is where you provide multiple examples to the AI model, and it is able to generalize to new examples of what you need.

·      One shot learning is similar, but instead of multiple examples, you only pass in one.

·      Zero-shot learning is where you directly pose a question type of machine learning without giving it any examples.

Case studies

Guidelines for Prompt Engineering in New Product Development (NPD) for a Digital X-ray System

1. Be Specific and Clear: Clearly define the stage or aspect of the Digital X-ray system development you are focusing on, such as ideation, system features, image processing algorithms, or cost optimization.

Example-

·      Generate ideas for improving the image resolution and reducing radiation exposure in a Digital X-ray system.

·      Provide competitive landscape for the digital x-ray with robotic arms in north America market.

2. Use Role-Playing: Position ChatGPT as a specific expert (e.g., radiologist, product manager, system engineer, or regulatory advisor) to get tailored advice for each phase of X-ray system development.

Example:

·      As a product manager, suggest key features that could enhance the customer acceptance of our Digital X-ray system.

3. Iterate for Depth: Iterate on broad responses with more focused questions to get refine features, technologies, or clinical applications.

Example: 

·      Iteration 1: "Can you provide more details on how AI can improve the detection of TB in Digital X-ray images?"

·      Iteration 2 : “ can you provide more case studies and efficacy of AI based image detection of TB in Digital X-ray system.

4. Incorporate Real-World Constraints: Include cost, regulatory, or technical constraints in your prompts to receive more realistic and implementable suggestions.

Example:

·      Propose  a system architecture to make a low-cost Digital X-ray system suitable for rural healthcare centers with limited power supply.

5. Use Comparative Prompts: Ask ChatGPT to compare different design approaches, technologies, or imaging modalities for better decision-making.

Example:

·      Compare the advantages and disadvantages of using flat-panel detectors versus charge-coupled devices (CCD) in Digital X-ray systems.

6.     Ask for Evaluation Metrics: Request ChatGPT to suggest key performance indicators (KPIs) that would help measure the success of your X-ray system.

Example:

·      What performance metrics should we focus on to assess the clinical effectiveness and image quality of our new Digital X-ray system?

7. Stimulate Innovation: Encourage creative problem-solving for existing challenges, such as improving image quality or integrating new technologies like AI or IoT.

Example:

·      What innovative solutions can we implement to improve the automatic calibration of our Digital X-ray system for various body types?

·      Describe the calibration steps and provide output based on the proposed solutions.

·      Compare the results with attached results based on the previous calibration for various body types

Example Prompts for Different Stages of NPD for a Digital X-ray System

1. Ideation and Concept Development

• Generate three innovative design ideas to make Digital X-ray systems more compact and portable without compromising image quality.

• What are potential features that could differentiate a new Digital X-ray system designed for use in emergency departments?

• What are the key risks associated with product market fit of the new digital X-ray system?

• Provide a competitor analysis of the current Digital X-ray systems in the market, focusing on image resolution and cost-efficiency.

• What are the key trends in regulatory requirements for Digital X-ray systems, especially in Europe and the U.S.?

• Outline the technical feasibility  study of using flat panel detector in the new digital X-Ray system?

2.  Product Design and Features

• List advanced image processing features that can enhance the usability of a Digital X-ray system in detecting microcalcifications.

• How can AI be incorporated into the X-ray system to automatically classify different types of bone fractures?

• How can the drive system more user friendly and efficient to reduce the work flow timing?

3. Prototyping and Testing

• What are some rapid prototyping methods we can use for developing a new detector array for our Digital X-ray system?

• Suggest testing strategies to validate the accuracy and consistency of our Digital X-ray system's image processing algorithms in detecting chest abnormalities.

4.  Product validation and verification

·      Please prepare a verification strategy based on the above product features

·      Prepare a complete work flow for the manufacturing, service and clinical validations

4. Product Launch and Go-to-Market Strategy

• What are the most effective go-to-market strategies for launching a Digital X-ray system designed for small clinics and healthcare providers?

• Develop a phased launch plan for introducing a new portable Digital X-ray system aimed at disaster relief efforts.

Case Study – Portable XRAY for emergency department for Hospital  and some standard questions. The answers restricted to 50 words only for this blog. ChatGPT is based on GPT-4 version.

Q	Creative setting	Prompt	Response	Remarks
	Temperature=1	Generate three innovative design ideas to make Digital X-ray systems more compact and portable without compromising image quality.	1.      Foldable Modular Design: Lightweight frames with plug-and-play detectors and generators enable portability and quick assembly. 2.      AI-Enhanced Imaging: Compact detectors paired with AI algorithms ensure high-quality imaging with reduced hardware. 3.      Wireless, Battery-Powered System: Advanced batteries and wireless connectivity support true portability, seamless image transfer, and field use.	Further deep dive into each ideas will through more information.
2		What are potential features that could differentiate a new Digital X-ray system designed for use in emergency departments?	A Digital X-ray system for emergency departments should offer rapid deployment, portability, and AI-driven diagnostics for swift, accurate results. Key features include bedside imaging, rugged durability, seamless wireless connectivity, and ergonomic controls. These ensure reliability, efficiency, and usability in critical care, enhancing patient outcomes and streamlining emergency workflows.
3		What are the key risks associated with product market fit of the new digital X-ray system?	Key risks include misaligned features with emergency department needs, high cost limiting adoption, inadequate image quality or reliability, and poor integration with existing healthcare workflows. Regulatory non-compliance and lack of differentiation from competitors may also hinder acceptance, impacting the product's ability to achieve market fit and commercial success.
4		Provide a competitor analysis of the current Digital X-ray systems in the market, focusing on image resolution and cost-efficiency.	Competitors like Siemens Healthineers, GE Healthcare, and Philips offer high-resolution Digital X-ray systems with advanced imaging software. Siemens leads in image quality with AI integration, while GE emphasizes cost-efficiency and portability. Philips balances innovation and affordability. Smaller players often provide budget-friendly options but may compromise on resolution and features.
5		What are the key trends in regulatory requirements for Digital X-ray systems, especially in Europe and the U.S.?	Key trends include stricter radiation safety standards, enhanced cybersecurity for medical devices, and adherence to EU MDR and FDA guidelines. Emphasis is on real-time data security, AI algorithm transparency, and patient dose monitoring. Regulatory focus on interoperability and quality management ensures compliance with evolving healthcare and technological advancements.
6		Outline the technical feasibility  study of using flat panel detector in the new digital X-Ray system?	A technical feasibility study for using flat panel detectors in digital X-ray systems for emergency rooms involves assessing image quality, radiation dose, power requirements, and integration with existing systems. Key factors include detector resolution, cost-effectiveness, portability, and robustness for handling emergency conditions, ensuring rapid imaging with minimal patient risk.
7		List advanced image processing features that can enhance the usability of a Digital X-ray system in detecting microcalcifications.	Advanced image processing features for detecting microcalcifications in Digital X-ray systems include high-resolution contrast enhancement, edge detection algorithms, AI-driven lesion segmentation, multi-scale image analysis, and noise reduction filters. These features improve visibility, accuracy, and sensitivity in identifying subtle microcalcifications, aiding early detection of breast cancer.
8		How can AI be incorporated into the X-ray system to automatically classify different types of bone fractures?	AI can be integrated into X-ray systems using deep learning models to analyze images and automatically classify bone fractures. Convolutional neural networks (CNNs) can detect and categorize fractures by type, such as simple, comminuted, or spiral, based on patterns and features learned from large datasets, ensuring fast, accurate diagnosis.
9		How can the drive system more user friendly and efficient to reduce the work flow timing?	The drive system can be made user-friendly by incorporating automated positioning, AI-assisted guidance, and ergonomic controls. A touch-screen interface with pre-set imaging protocols reduces manual adjustments, while wireless connectivity enables quick image transfer, reducing overall workflow time and ensuring fast response in emergency settings.
10		What are some rapid prototyping methods we can use for developing a new detector array for our Digital X-ray system?	Methods like 3D printing, CNC machining, and MEMS (Microelectromechanical systems) fabrication allow quick iteration of detector arrays. Using simulation software to model X-ray interactions and materials testing can accelerate development, ensuring the detector array meets performance, size, and cost requirements for a portable Digital X-ray system.
11		Suggest testing strategies to validate the accuracy and consistency of our Digital X-ray system's image processing algorithms in detecting chest abnormalities.	To validate image processing algorithms, conduct benchmark testing using standardized chest X-ray datasets. Implement cross-validation with expert radiologists, phantom testing with synthetic abnormalities, and real-world clinical trials. Regular comparison with existing systems ensures accuracy and consistency, crucial for detecting chest abnormalities in high-pressure emergency scenarios.
12		What are the most effective go-to-market strategies for launching a Digital X-ray system designed for small clinics and healthcare providers?	Target small clinics with cost-effective, portable Digital X-ray systems through direct outreach and training programs. Leverage partnerships with distributors to ensure quick regional availability. Emphasize affordability, ease of use, and low maintenance costs, and offer financing options to reduce upfront investment for smaller healthcare providers.	Some of these responses are generic.
13		Develop a phased launch plan for introducing a new portable Digital X-ray system aimed at disaster relief efforts.	Start with a pilot deployment in one region, focusing on fast setup and ease of transport. Phase 2 involves broader deployment, providing training to field teams and integrating the system into disaster relief protocols. Phase 3 expands to global humanitarian partners, offering ongoing support and gathering feedback to refine future deployments.

The answers provided above are generated by GPT. While some responses may be more general and could fall short of the expertise required in this domain, a deeper exploration of each answer could yield more precise insights. It's important to ensure proper summarization, as key details might be missed in broader responses.