ConductorAI

Trying to find answers to specific questions can be a hassle, you can go on an endless loop of questions and answers, and get nothing. The ConductorAI or CAI for short helps you avoid this issue.

Comment

Inspiration

We have conducted a small case study with no tech-savvy subjects (our grandparents) and tasked them with the goal of finding specific bits of information from the NJ Transit FAQ web page. It took them an average of 3 and a half minutes to actually find the information. With that information, we thought to ourselves that it was crazy that it took them that long. We tried it ourselves, and it even took us a decent amount of time. There is just too much information on that page and it looks cluttered. With the rise of AI, we decided to take a crack at revamping the NJ Transit FAQ page, with an AI helpdesk chatbot, CAI.

What it does

CAI is an Ai helpdesk chatbot that assists people with any questions that they might have about NJ Transit. The chatbot utilizes natural language processing to query through the frequently asked questions dataset provided by NJ Transit to quickly return an accurate answer. It strips away the hassle from its users of having to go onto the FAQ webpage to manually search for their answers.

How we built it

CAI was made with Python, and some of the libraries include Tensorflow, TFLearn, JSON, Numpy, and NLTK. We took the NJ transit FAQ JSON file and refactored it to suit our needs. We separated each question and answer into tags, patterns (possible question inputs), and responses.

Challenges we ran into

The dataset provided by NJ Transit wasn't in a viable format for us to utilize, so we have to refactor it to suit our purposes. We also ran into an issue with getting a viable tagging system set up for AI.

Accomplishments that we're proud of

This was our first attempt at creating an AI chatbot, and we even managed to get a working product. We were able to be more specific with the tags and have CAI respond with the correct answer corresponding to the correct question.

What we learned

We learned the basics behind what it takes to create an AI bot along with the training process. We also learned a lot about how natural language processing works, manipulating JSON files, and even a great deal about Tensorflow and training models.

What's next for ConductorAI

Finding wicked problems that challenge the bot to find a simple solution for the user and hopefully make some connections with NJ Transit.

Built With

Share this project:

Updates