VioletIQ

Inspiration

One of the biggest reasons why gender stereotypes persist is that they are normalized in areas ranging from family to media and education. If you read a news article about a company, then the CEO is likely to be a man. Similarly in movies, the romantic interest of the main character is typically a woman, whom the main character “earns” or saves at the end of the movie.

What it does

VioletIQ provides tools to stop the perpetuation of gender stereotypes through education. More specifically, VioletIQ is a toolkit to interact with important information with bias detection, summarization, link detection, semantic query answering, and generated flashcards.

How we built it

We built it using GPT-Index, a python framework that embeds text to build an indexing data structure that can then be used to help answer semantic queries. For example. If you had the query “What is the main them of the book” this would fail with a keyword search, but with GPT-Index we can answer this question in a contextual way. For some of the models, we used few shot learning instead of indexing to get the response in the desired form. Additionally, we used Diffbot to do web-scraping because it can scrape a wide variety of different websites and file formats.and is modularized in an easily accessible api. Lastly, we made use of streamlit to build UI and deploy an app with little to no overhead.

Challenges we ran into

Latency: One challenge we ran into was latency because indexing is an extremely computationally intensive task. As a result, we made use of the caching and memoization in streamlit to get our app working at peak performance.
Data: It was hard to find datasets for the bias detector sexism. There were a couple that invovled Given the time constraints and lack of data, we decided to go for GPT-Indexing and few shot learning.
History: Another challenge we ran into was storing a history for queries because we had to maintain persistent session variables to store this data.

Accomplishments that we're proud of

Integrated many different tools within the toolkit
We believe that our project helps mitigate an impactful issue

What we learned

We learned that instead of focusing on extremely technically difficult projects and over-engineering solutions. We could easily deploy a consumer-facing application with minimal effort by leveraging existing technologies such as API’s and hosting platforms. We learned that simplicity results in projects that run smoothly

What's next for VioletIQ

We want to create a bias checker for other forms of media such as photos and videos
We hope to build more infrastructure and reporting tools to allow us to better track and understand our customers, so we can provide tooling to fulfill their specific data needs