Tweet Mimic

Try it out!

Each tweet that you generate is automatically uploaded to our twitter page. Follow us for more!

Pick an account, enter a basic prompt and click submit!


Thank you! Your message has been sent.
Unable to send your message. Please fix errors then try again.

Our Process

Getting Tweet API & Datasets
a. In order to start training our models, we needed large datasets of tweets from each of the personalities we sought to mimic. We had two methods of getting the data: prepare datasets and scraping tweets from Twitter’s API.
b. On the latter, we had to build a program that would scrape tweets from Twitter’s API using a developer account and tweepy. This was a new challenge for us -- to build a tweet scraper to compile data.

1 

2

Cleaning data
a. While cleaning the data was more straightforward for the preprepared datasets, the cleaning of the raw tweet data posed a greater challenge. One unique problem we encountered in analyzing the tweet dataset was how to remove the quote links.
b. We also removed other non-content-related aspects of the tweet, such as whitespace, symbols, and rarely used characters that might have confused the training model.
Training Model
a. For each personality, we started with EleutherAI's 125 million parameter GPT-neo model.
b. Each personality had a separate model that was trained specifically on tweets from that personality. This allowed us to create several models, each trained to emulate one personality.

3

4

Designing and Integrating the Website
a. To turn the backend steps into a user-friendly product, we had to ensure the website's JavaScript, HTML, and CSS was compatible with the Python-based models. This presented some new challenges, as we had to manipulate the JavaScript despite lacking relevant coding experience in that language.
b. We also had to iron out the details of the website that had been overlooked upon initial creation. Through a cumulative effort, we were able to fix the errors and address the minor complexities of integrating 8 unique models into the website.

5

Completion & Presentation
a. Following the completion of the integration, our project was only missing a few components. First, we needed to run a quality check to search for potential uncaught errors or bugs in the functionality of the website or code.
b. Then we created a presentation documenting our project, it's functionality and purpose, and the steps we took to achieve our goal. Throughout this project we developed and enhanced our collaborative, problem-solving, and creative skills ultimately gaining real hands on experience.