DinoAura | Devpost

Start page
Quiz
Result page
Chat

Project story

Inspiration

When we are given advice by others, we typically do not take it because we do not internally align with it. Furthermore, when searching for answers online, we often see generalized advice that may not fit our specific needs. Therefore, in order to generate sound advice that the user will actually take, we developed an emotionally intelligent chatbot that will emulate the user personality and give advice that internally aligns with the user’s personality, creating personalized guidance that the user is likely to embrace.

What it does

After completing a personality quiz, users will get classified as one of 5 personality types: Lemon Dino (The Relaxed Minimalist), Acai Dino (The Easygoing Free Spirit), Kiwi Dino (The Balanced Explorer), Taro Dino (The Thoughtful Realist), and Lychee Dino (The Cautious Introvert). Then, they are able to chat with our DinoChat, which will emulate their personality type to give them personalized advice that they are more likely to implement. By implementing a quiz, users are able to reflect on themselves and learn more about their personalities.

How we built it

Here was our basic timeline and breakdown of our project

Create Figma designs for the homepage, the quiz, the results page and the chat bot
Create a personality test and a list of questions that we can use to ask our users
Using React create a simple quiz and implement our questions
Create the homepage and the results page
Redesign the three pages to match our Figma designs
Implement a chatbot from Deep Chat
Connect the Cohere API chat tool to our chatbot
Connect the responses from the quiz and their individual descriptions to our chat’s “system_prompt”
Created a testing set for our personality classification model, trained our classification model using Cohere, and ran our DinoChat responses in the classification model to determine how successful our - DinoChat was in replicating the personality of the user via a confidence level form 0 - 1.

Challenges we ran into

Some challenges that we ran into were that the confidence level that was generated by our classification model was not very promising, meaning that our chatbot was not extremely successful in replicating the user personality when giving out advice. Therefore, we need to continue refining our personalities so that the chatbot is able to identify the distinction between the personalities and give out differing advices depending on the personality it is supposed to emulate.

Accomplishments that we're proud of

We are proud of the user interface. The user interface has an aesthetic design, is simple and is easy to use. The chat bot is easy to use and provides insightful advice, which was the goal of the project. Additionally, we’re proud of being able to come up with cute mascots (Lemon Dino, Lychee Dino, etc.) to represent each personality, and be able to design them in an appealing way.

What we learned

We learned a lot about how to implement a lot of existing useful code and various tools to create something innovative. For example, instead of recreating the chat-bot we found a way to implement Deep Chat’s chatbot which was a lot easier and gave us a lot more time to work on advancing our AI tool in Cohere.

Additionally, we were able to break down the problem before getting started so that our code didn’t overlap and we were able to work together very well without interfering with one another. We had a clear plan and clear timeline that allowed us to finish this project. From our last hackathon we were unable to fully finish our idea because we didn’t have a clear plan and were working not as seamlessly.

What's next for DinoAura

We generated a test set of model responses per personality type and used them to train our classification model using Cohere API. Then, we used DinoAura and classified the responses that DinoChat gave us using our trained classification model, which yielded a confidence level for each personality type. This helped us see how successful the chatbot was at emulating the personalities that the user was assigned. Since our confidence rates are not at an ideal threshold, the next steps of DinoAura would be to continue refining our personality types so that DinoChat can better emulate the user's personality type and therefore, give more feasible advice that the user will implement.