RolePlay Trainer

• 

  Architecture Diagram

Inspiration

A friend of mine who is teaching Vets asked me to build a system that helps his student handle future patient interaction, we figured it's possible to use gen ai to do this

What it does

for MVP it will load a pre-created script, along with character design and situation, goal to achieve, and have the LLM improvise with user to recreate situations we want to train students on. After the interaction, an evaluation agent will provide feedback with actionable items to practice more or improve. Both the role playing part and the evaluation part works with both English and Traditional Chinese.

How we built it

a python fastapi backend with agents built using agent development kit, and a vue.js frontend

Challenges we ran into

adk and chat and fastapi doesn't have a well documented way of working together python's packaging structure, and how I structure fastapi is not as intuitive as I wanted them to be

Accomplishments that we're proud of

We established a way to have agents run and served out of a custom FastAPI server, as well as keeping adk run / `adk web' workflow still working with what adk expects in terms of agent source file location

What we learned

ADK is opinionated and in some sense not as intuitive to use if your workflow relies heavily on structured output.

What's next for RolePlay Trainer

A design feature that can create new characters, scenarios as well as actual turn by turn script to create more customized training And enable websocket for using live voice to make training more immersive.

Built With

• adk
• fastapi
• python
• vue