AI Avatars You Can Talk To. And They Talk Back.

What if learners could speak to an avatar right inside Articulate Storyline? That’s what this demo explores.

We’ve written previously about our use of ‘standard’ AI avatars in eLearning, but this is something different. This avatar listens to you. Not only that, but it talks back.

Meet Phil

He’s an “experienced eLearning designer and developer with strong instructional design skills, a rare all-rounder who understands visual design, learning theory, development, and strategy, and he’s great at blending all of them. He doesn’t just make courses, he crafts learning experiences that look good, feel right, and actually work”, at least that’s the persona we’ve given him!

Ask Phil something about eLearning and he’ll do his best to answer. Phil uses GPT-4o mini as his Large Language Model (LLM) for generating responses in real-time. He connects to this LLM through his API, allowing him to engage in dynamic and interactive conversations. We’ve also fed him with some URLs too. These help him with anything more specific that we might want him to know.

When answering questions, we’ve asked Phil to “Be conversational: Talk like you’re chatting with a colleague over coffee. Casual but not sloppy. Include emotion and wit: Be warm, a bit cheeky at times, and very human”.

On top of this he’s also a polyglot, which means he could prove very useful when localising eLearning.

Yes, It Runs Inside Storyline

As a lot of our eLearning is developed in Articulate Storyline, it made sense to embed our avatar into a Storyline project.

To do this we generated our avatar in HeyGen, then created an HTML file that we embedded in Storyline using a web object.

At first when we tested it locally, we hit a snag. The microphone didn’t work so Phil couldn’t hear us. After a bit of digging, we found that we needed to serve the HTML file over HTTPS for the browser to allow microphone access. So we uploaded the file to our server, and … it still didn’t work!

It turns out that by having the HTML file on our server, and the Storyline files hosted locally we’d created a cross-domain issue.

Thankfully that was an easy fix and uploading the Storyline files to the same location as the HTML file solved the problem.

Some Things to Keep in Mind

For the most part, this all works well. Phil is a little enthusiastic at times and in our early attempts he seemed to shout ‘Hey!’ at the start of every sentence. He was also quite fond of telling us his name every couple of minutes too.

As with all generative AI, crafting the right prompt is essential, and we think we fixed Phil by adding the following:

• Introduce yourself once at the start of the chat. Only mention your name once as it is a bit strange to introduce yourself more than once to the same person.
• Don’t be over-enthusiastic by calling out words like “Hey!” at the start of a sentence. Just be calm.

If you have a conversation with Phil, I’m pretty sure he’ll display plenty of quirks we haven’t experienced yet, but that’s what makes him human-ish.

The Possibilities: Looking Ahead

I wouldn’t position this as client-ready just yet. It’s still a bit rough around the edges, but the potential it shows is genuinely exciting.

Once Phil and his peers prove themselves to be trustworthy teammates, I can see them becoming very useful in an eLearning context.

Much of our work centres on pharma sales, and with the right knowledge base, I can see interactive avatars becoming powerful coaching tools for sales teams. Imagine learners practising objection handling in real-time, or building interview confidence through natural conversation. The opportunity to add real value to the learning experience is huge.

Watch the video below to hear Phil answer our questions about eLearning

Related eLearning Posts