PirateShield: Automated Copyright Infringement Detection

Inspiration

Viral content, particularly copyrighted material and deepfakes, has huge potential to be widely proliferated with Generative AI. This impacts artists, creators and businesses, as for example, copyright infringement causes $11.5 billion in lost profits within the film industry annually.

As students who regularly come across copyrighted material on social media, we know that manual reporting by users is clearly ineffective, and this problem lends itself well to the abilities of AI agents. A current solution by companies is to employ people to search for and remove content, which is time consuming and expensive. We are keen to leverage automatic detection through our software, and also serve individuals and businesses.

What it does

PirateShield is a SaaS solution that automatically finds and detects videos that infringe a copyright owned by a user. We deploy AI agents to search online and flag content using semantic search. We also build agents to scrape this content and classify whether it is pirated, using comparisons to copyright licenses on Youtube. Our prototype focuses on the TikTok platform.

How we built it:

Our platform includes AI agents built on Fetch.ai to perform automatic search and classification. This is split into a retrieval feature with semantic search, and a video classification feature. Our database is built with MongoDB to store videos and search queries.

Our frontend uses data visualisation to provide an analytics dashboard for the rate of True Positive classifications over time, as well as rates of video removal.

Challenges we ran into

We initially considered many features for our platform, and had to distill this into a set of core prototype features. We were also initially unsure how we would implement the classification feature before deciding on using Youtube's database. Moreover, testing our agents end-to-end on queries involved much debugging!

Accomplishments that we're proud of

As a team, we are proud of identifying this impactful problem to work on, and coordinating to implement a solution while meeting for the first time! In particular, we are proud of successfully building AI agents to search for and download videos, as well as classify them. We're excited to get our first users and deploy the remaining features of the platform.

What we learned

Our tools used were Fetch.AI, Google APIs, fast RAG and MongoDB. We upskilled quickly in these frameworks, and also gained a lot from the advice of mentors and workshop speakers.