I was hired as an instructional designer to create storyboards for a range of topics. We were working in an Agile methodology to mass-produce scripts in time for studio recording. I found a way to leverage AI and automate the process by using different LLM models for specific tasks, generating output that provided a strong foundation for the scripting process.
The company was using Agile methodology to produce multiple courses on technical topics, including blockchain technology and cryptocurrency. They had tight quotas and a large volume of content to complete
Originally I had created a custom GPT to produce a draft script. However, I found that I had to continuously click from Chat GPT to claude, to perplexity. I realised with strong enough and effective prompts and using the openAI playgrounds vector database I could automate the process
Creating the meta prompts meant fusing my expertise in prompt engineering and instructional design. For example, applying the personlisation principle,active voice, talking directly to the listner etc. Additionally, recording my natural voice rhythm and using this in the prompt as an example, meant the output got closer to a naturally sounding script. Anyone who has used Chat GPT without adding in few-shot prompting knows it will sound like a robot. Then we enter the uncanny valley.
I created a specialised assistant in the Openai playground to draft the outline and structure of the script. Then this was handed to Claude, which excels at writing scripts. Finally, a bonus module which I find is sometimes overlooked is setting up a quality assurance AI with a light weight model to check against a QA list. These comments can be checked and then worked with further in Claude before adding the human touch.
It was pieced together using make.com with google docs with the final output a one off or series of scripts on a topic. Below you'll see the quality of the first draft.
After analysing the the amount of scripts we could produce in a day, with the CLO he said it was the equivalent of two instructional and a senior designer. This amassed to saving $2,500 per week.
The scripts produced were not the complete polished example, however, it provided myself with a powerful starting point with the editing and saved an additional three hours of writing time.