Update: I'm creating a simple HTML page that demonstrates the model in action.
This is a repository for my Prompt Architecture Model paper. This work orginated with tha AIQL that I created a while ago. I got a lot of good feedback so decided to keep pushing in the same direction. I ended up with quite a comprehensive tiles or canvas model which I called Prompt Architecture Model. I received a lot of really good feedback (some positive and some negative, but all useful) with my last work and look forward to any challenges, comments or criticisms you may offer.
Here's the abstract from the paper.
Prompt engineering plays an important role in optimizing interactions with large lan-
guage models (LLMs), facilitating task-specific guidance through structured inputs. For instance,
in customer support, a prompt can guide an LLM to provide clear and concise responses tailored
to user queries, ensuring efficiency and relevance in its output. Despite its importance, the field
remains fragmented, with practitioners often relying on trial-and-error approaches and a lack of
standardized methodologies. This paper introduces the Prompt Architecture Model (PAM), a
structured framework that formalizes prompt design by providing a systematic approach. Unlike
existing frameworks, PAM integrates modular components such as objectives, context, and val-
idation into a cohesive structure, ensuring consistency and adaptability across tasks. Drawing
on principles from modular design and AI workflows, PAM organizes the prompt creation pro-
cess into ten interrelated components, including task objectives, audience specification, context,
sub-tasks, and validation metrics. This approach enhances clarity, consistency, and scalability,
addressing key challenges in the field. Through practical examples, the paper demonstrates
PAM’s adaptability across various domains, such as academic research, business analytics, and
content creation. While the framework shows notable promise, the paper also discusses its lim-
itations, including its current focus on text-based tasks and potential challenges in multimodal
applications. Future research directions include expanding PAM to multimodal tasks, automat-
ing prompt generation, and integrating the framework into existing AI tools for broader adoption.
PAM contributes to the growing body of work in prompt engineering, offering a replicable and
scalable methodology that improves the effectiveness, efficiency, and ethical robustness of AI
interactions.