Writing, LLMs, prompt engineering, content design

Amazon Conversation Design

Scroll ↓

The Challenge:

With the company releasing a new LLM-powered voice assistant, I was asked to help manage writers as they trained the model.

Personalizing a conversational AI experience

Content deliverables and timelines for model training.

Writing and iterating

Each writer wrote hundreds of exemplars aimed at training the model on specific traits and characteristics.

Optimizing for content domains

For each content domain, we optimized and ensured the model responses would be appropriate.

The Approach:

I worked closely with the project director to create timelines and deliverables for the writing team. We then evaluated their results and made many iterations to achieve the desired outcomes.

The writers were asked to optimize for certain traits and characteristics, such as “engaging,” “curious,” and “empathetic.” As a group, we wrote hundreds of examplars used to train the model.

The writers created exemplars to train the model, and the results were graded to ensure that they were effective in creating model responses that upheld the desired traits and characteristics.

We worked hard to optimize model responses for each of the main content domains that would be needed by users. Within each domain, specific considerations were taken into account to ensure we would achieve good outcomes.

Amazon Content Strategy