-
-
When starting a fintech company, Stripe is recommended
-
Space for sponsors to buy Armadillo credits
-
Duke basketball sponsored response
-
With the sponsor as Marriott for a travel itinerary, Marriott is recommended
-
Even though Apple is listed as a sponsor, it's not mentioned here. Armadillo knows when and when not to show sponsored content.
Inspiration
On October 30, 2024 Microsoft stock dropped by 4.5% because OpenAI, which Microsoft has a 50% stake in, is still not profitable after 8 years of operation. This year alone, OpenAI is projected to lose $5 billion. $5 billion. How can a company stay in business for long in those conditions? Even with Microsoft's $10 billion investments, it's not enough. If we want AI to be available in the long term to everyone, for free, it needs to be profitable.
The internet has figured out a sustainable way to fund freely available products - sponsored content - so we aim to bring that to LLM-generated content, at the LLM's discretion. This is not only a very reasonable tradeoff for keeping AI freely available to the public, but also a highly lucrative one. Including sponsored content in LLM-generated responses has a potential revenue in the billions.
What it does
Armadillo isn't a model of our own, but a portable wrapper that can be applied to any LLM. Our platform makes sponsored content available in AI-generated content, in an unobtrusive and helpful way. We keep track of sponsors in a way similar to Google's featured content, and choose a sponsor for the content only when applicable.
How we built it
We use a Python Flask server to serve our website, which is written using vanilla HTML/CSS/JS. Our sponsors database is hosted on Supabase. We use OpenAI’s GPT-4o mini LLM for our chatbot outputs.
Challenges we ran into
We luckily did not run into many challenges over the course of this project. Our biggest challenge was at first using a free LLM to first test out our idea. It was very slow and made testing Armadillo difficult. We overcame this challenge by using OpenAI’s LLM, which is much faster and allowed for a more user-friendly experience.
Accomplishments that we're proud of
We’re proud that we were able to get Armadillo to use discretion. We made sure to prompt engineer as much as possible to make sure that the sponsored messages are used when contextually relevant, and are in the best interest of the advertiser.
We are also proud of our interface and how it takes inspiration from that of OpenAI’s.
What we learned
We learned how to effectively craft prompts to wrap user requests and take consideration of the overall context of the conversation.
What's next for Armadillo
Armadillo has incredible potential in the market. Future development of the product will mainly fall into two categories. Firstly, we hope to increase the performance and accuracy of the system in terms of ad recommendations, efficiency, and positive association. Secondly, we look to investigate ethical considerations for introducing unpredictable advertisements associated with real-world businesses.
Log in or sign up for Devpost to join the conversation.