Writeeasy | Devpost

Inspiration

Have you been in a situation where your friend asks about that fascinating movie you just watched yet you find yourself get stuck and don't know how to describe it? All of us have.

As our team attended the Workshop presented by Cohere, we were incredibly impressed with the AI's capability with generating text-based media. We decided to use the technology to help alleviate stress in menial writing tasks, such as describing a movie to a friend, making a Twitter post about the team introducing a place to others, etc.

What it does

Our product is a front-end web platform serving anyone who needs a piece of paragraph on a simple everyday topic.

The functionality of Writeeasy is very simple. Given a prompt(in question format) or a list of keywords, our web app will use a Cohere model to generate a response of up to/around 80 words.

For example, a response to the prompt: Why do you like coding? generated by our software is:

Coding has opened my eyes to a new world of possibilities. I used to think coding was about writing computer programs. Now, I see coding as a way to make something new. It is about creating a new idea, and seeing it come to life. Coding is a way to create something that had never existed before.

The response for personal questions, such as when your newly-made friend is curious about why you like a particular hobby ("Why do you like playing basketball?"), may be hard to answer on the spot. Our goal is for our AI to save user time and effort if they ever find themselves get stuck. They can look at what our AI writes, and use is either as an inspiration or as a template based on which further edit is added. Essentially, we are providing the user with a starting point and then the user ample freedom to build on that, hopefully with more ease.

How we built it

We built the project entirely using Cohere, an open-sourced AI model that provides many NLP functionalities. We split the workload into two different areas, mainly the backend and the frontend. In the backend, we made API calls to Cohere. Then we tested the model with our desired prompts. Afterwards, we customized our prompt successfully with prompt engineering, effectively using 3 distinct prompt answer examples to guide the AI on the right path.

In terms of the frontend, we used the Python-based web framework called Flask. The website is hosted by Heroku.

Challenges we ran into

With just one prompt, we get a variety of different results. Sometimes, rather than getting the AI answering the prompt, the AI makes many new prompts. Sometimes we get the AI asking its own similar prompts, rather than answering the prompt itself. It took us a long time to be able to figure out how to get the AI to output the specific response we want, which is a reasonable answer to the prompt. We also had many setbacks in relation to getting the Cohere platform working, such as setting up Client key, using stop phrases, and others. Luckily, we managed to get Cohere working successfully, as demonstrated by our implementation.

In terms of the frontend, we were met with certain difficulties when it comes to font and formatting. However, we were able to pull it off at the end.

Accomplishments that we're proud of

We were able to collaborate as a team even as we previously didn't know each other. Writeeasy still has plenty of limitations, but we were able to successfully apply an AI NLP generative model in a front-based web-interface despite having no experience with this area.