I was inspired by AI to talk with fictional characters, and modern innovations in deep learning AI.

What it does

Allows you to talk to a semi-realistic bot of Gandalf from Lord of the Rings.

How we built it

I used a deep learning NLP model and trained it on various (but limited) kinds of data.

Challenges we ran into

Deep learning AI is very computationally intensive. You need a good GPU to train large models, and sometimes you run into issues with the model you trained for a long period of time, which wastes a lot of time.

The model itself is very imperfect, and it does not nearly cover all user input. It does, however, catch a good amount of input of what you would expect the average person to know about Gandalf.

The benefit of NLP modes is that you can cover a lot of input with a small amount of data. The downside is that you can have very undefined behavior if your model lacks training data for a specific input.

Accomplishments that we're proud of

I built an NLP model for Gandalf and included an actions server to handle additional logic for when the user asks about Gandalf's friends and enemies. The actions server is cool because it allows you to implement very complex logic to work side-by-side the NLP model.

What we learned

Deep learning is imperfect without a ton of data.

What's next for Gandalf AI Chatbot

Add more training data to see how good it can get, and add more logic to the actions server to add additional realism to the model.

Share this project: