Redcommend

Inspiration

After long hours on Reddit, we wanted to build a tool that allows people to reduce the amount of time it takes to find the answers they are searching for.

What it does

Redcommend allows users to enter a topic and a subreddit of their choice, and it produces a synthesized summary of discussions about the given topic. While users can technically enter any topic, our project is currently structured to provide most accurate results for movies and TV shows.

How we built it

Once the user has inputted a movie name and the target subreddit, we use the PRAW wrapper for the Reddit API to gather the top comments on discussions about the movie in the given subreddit. We then used the Classify API from Cohere to deduce common tones (positive, negative, neutral) and themes (characters, soundtracks, etc), as well as the Summarize API to summarize discussions about the topic.

Tools used:

PRAW wrapper for Reddit API
Cohere API
Figma
Flask

Challenges we ran into

One challenge we have faced is a slow response time, as our backend calls multiple external API's that need to complete time-intensive tasks such as web-scraping and text generation. We are hoping to resolve this by implementing chained API calls for different processes.

Accomplishments that we're proud of

We are proud of learning to use new tools and technologies in a short time period to make a creative project that has many possibilities for extensions.

What we learned

We gained a much better understanding of how to work with AI tools such as Cohere. Namely, we learned how to structure our input parameters to get the most relevant output for our purpose. We also learnt how to use Figma to create UI prototypes.

What's next for Redcommend

We want to introduce other types of content that subreddits can recommend to users such as books, restaurants, class courses, etc. At the moment, our project works best with movies and TV shows, as our classification examples for the Cohere 'Classify' API are centered around that. However, we are hoping to implement functionality so that these examples can be self-generating, allowing the scope of this project to widen.