AI & ML Patent Resources

The Refuser Pattern

Prompt Pattern

Free

The Refuser Pattern is a prompt pattern used in AI and large language model systems. Explicitly defining negative constraints or unacceptable output types in the prompt. The core purpose of this pattern is to provide a repeatable structure that developers, researchers, students, and AI application designers can apply when building or improving interactions with language models. In practical use, it is especially relevant to safety filters, avoiding specific styles, legal compliance. Rather than treating every AI request or workflow as an isolated implementation, this pattern provides a recognizable approach that can be adapted to different applications, models, domains, and levels of system complexity. It is categorized as beginner-friendly, making it suitable for users who are learning how structured AI interaction and reusable design techniques work. Implementation details can vary depending on the selected model, prompt structure, available tools, external data sources, orchestration framework, and application requirements. When applying the pattern, users should define the intended objective, inputs, expected outputs, constraints, validation method, and failure-handling behavior so that the resulting workflow remains understandable and testable. The associated dataset tags are prompt,constraint,safety, which can be used to group and discover this pattern alongside related techniques. The source included with this record should be consulted for the original reference or additional technical context, while real-world implementations should be evaluated with representative tasks before production deployment.

Fine-Tuning Workflow (LoRA)

Workflow

Free

Fine-Tuning Workflow (LoRA) is a workflow used in AI and large language model systems. The process of lightweight fine-tuning a base model on custom data using parameter-efficient methods. The core purpose of this pattern is to provide a repeatable structure that developers, researchers, students, and AI application designers can apply when building or improving interactions with language models. In practical use, it is especially relevant to customization, domain specialization, cost efficiency. Rather than treating every AI request or workflow as an isolated implementation, this pattern provides a recognizable approach that can be adapted to different applications, models, domains, and levels of system complexity. It is categorized at an intermediate level and is most useful once the basic concepts of prompting, LLM applications, or agent workflows are understood. Implementation details can vary depending on the selected model, prompt structure, available tools, external data sources, orchestration framework, and application requirements. When applying the pattern, users should define the intended objective, inputs, expected outputs, constraints, validation method, and failure-handling behavior so that the resulting workflow remains understandable and testable. The associated dataset tags are workflow,training,mlops, which can be used to group and discover this pattern alongside related techniques. The source included with this record should be consulted for the original reference or additional technical context, while real-world implementations should be evaluated with representative tasks before production deployment.