Discard

Discard, the bot that stops spam, fraud and cyberbullying in its tracks. Discard makes use of a computer vision model and LLM to scan and block suspicious messages.

Comment 3

Inspiration

Moderation on online platforms like Discord is an enormously difficult task. Moderators have to sift through hundreds or even thousands of messages a day. Many Discord servers have a shortage of moderators or even none at all. Some content moderators see can be very traumatizing, and many moderators end up with PTSD.

As teenagers, we’ve all witnessed cyberbullying before. 42% of children and teens have been cyberbullied, and 90% have witnessed cyberbullying. Children and teens who have experienced cyberbullying are four times more likely to have suicidal thoughts.

We also all have a friend or loved one who’s been the victim of a scam. Scams target the most vulnerable members of our society: young people, older adults and financially distressed people. Each year, the world loses 1 trillion to scams, many of which are online.

When we first became aware of all these problems, we realized we could create meaningful change on a platform that is used by us and 614 million other people, Discord. We came up with Discard, a Discord bot and the protection that works for you.

What it does

Discard automagically moderates your discord server. It leverages deep learning to automatically detect a malicious message. Malicious messages include spam, cyberbullying, harassment, inappropriate content and trolling. If it is determined to be malicious, it will then give the sender a captcha to stop automated spam messages. They are timed out if they do not complete this captcha within 60 seconds. Malicious messages are logged to an admin panel (website), where an admin can manually decide how to respond and see why the message was flagged as malicious.

How we built it

We used Git and GitHub for version control. We used HTML, CSS and JS for the dashboard and Discord login for dashboard accounts. We used Python, specifically Flask and Django for the backend. We used a ViT (CV model) to screen images in messages and an LLM (Llama-3) to screen text in messages. We used the VirusTotal API to detect malicious EXEs. We used Cloudflare Turnstile for our captcha. The services are hosted on our own servers, all running Debian.

Challenges we ran into

We encountered many challenges during the creation of this project, such as communication between devices for the API servers. Since most of us have protected our hardware with firewalls and other protection software, we were initially unable to do so. However, we were able to solve this issue by using a virtual private network, which we used to create a personal local network.

Additionally, our api endpoints received a ddos attack halfway through development and halted development for some time, as cloudflare sent captcha challenges to every api request that happened.

Documentation was also sparse when working with yolov8 object bounding boxes models. This resulted in many hours of experimentation.

Accomplishments that we're proud of

We’re proud we managed to implement almost everything we originally planned. We’re proud we made our prototype work, almost as much as our final product will be. We’re proud of the amazing video William came up with and filmed. Most of all, we’re just proud we survived our first hackathon.

What we learned

Since this was our first hackathon for all of us, it was a huge learning experience. We learned how to work on a lot of things in a very short amount of time. We learned that splitting up tasks could help us get more stuff done. We learned that taking breaks after long periods of work is very important. We learned how to deal with conflict when we had different ideas

Some other stuff we learned include: How to make proper API calls How to use Cloudflare Turnstile More Flask and Django knowledge How to use the Discord API How to convert Figma to HTML and CSS

What's next for Discard

Our goal is to make the internet a safer place, starting with Discord.

We’ve started with Discord but we don’t plan on stopping here. Discard can be ported to social media platforms like Instagram and communication apps like Whatsapp. In fact, we can port it to any platform with an API.

We also want to add new features like checking links (scraping websites that are linked to) for suspicious content, OCR as well as live video/audio detection. For non-spam content, we need to implement other protections because a captcha won’t work.

We want to make Discard available to everyone while sustaining our development costs. To do this, we’ll monetize Discard with a freemium model. Discard will be open source, and anyone can use it for non-commercial purposes. However, they’ll have to host all of the APIs themselves. We will provide a subscription model that is fully hosted and provides enterprise-level support.

To market Discard, we will use a referral-based model where customers get discounts if they refer Discard to other people. To scale up our servers, we will need $100,000 from investors. If we can get 1% of Discord admins to adopt the paid version, we can make 2 million dollars per year in revenue and a few hundred thousand in profit that we can use to scale up our servers.

Built With

Share this project:

Updates