GemAds

Inspiration

Ads have the potential to be used in AI chats. We wanted a way for advertisers to reach users inside Gemini answers due to the huge user base that they currently have while keeping payments on-chain. GemAds connects sellers to buyers and can show users targeted products that they truly need. The future of ads is in chatbots and we want to validate this idea.

What it does

Advertisers pay in SOL to show contextual ads in AI responses. Ads are matched to user queries via vector similarity and injected into Gemini output. A Solana escrow holds funds and routes them by outcome: money sent to Gemini on click, refunds when users don’t engage. Pricing scales with usage, including a token-based model, as well as a fixed amount that refunds. The amount of advertisement is based on the amount of SOL they pay. The user can talk with the chatbot and get very specific and personalized recommendations for products/services to use based on their chat history.

How we built it

Flask backend, Solana for payments, and JSON stores for advertisers and metrics. The middle-man escrow receives company SOL, handles refunds, and pays out to the Gemini Wallet. We integrated the Gemini/Supabase flow and exposed endpoints for the main backend (payment-received, refund-needed, transfer-to-gemini) so it can plug into a Flutter frontend. Web scraping was used to get information on a particular company and update the Supabase db. Eleven Labs was used for voice generation for easier use and cooler way to present output. .

Challenges we ran into

Many sites block web-scrapping so we had to create a fallback that searches in Gemini itself. When recording audio, Flutter has a record package that only records in a specific file type, which required converting to another filetype that ElevenLabs could process. Open Router and Digital Ocean credits were not available so having multiple LLM API keys was made difficult and database management required more tools.

Accomplishments that we're proud of

Voice and text initialization alongside a chatbot functionality was a big milestone in the project. Simulating an environment where user can interact in real time with professional responses. The ads at the end are implemented in a non-invasive manner while also promoting and encouraging users to buy the best product for them.

What we learned

Setting up multiple API endpoints especially without a centralized system like Digital Ocean to manage the entire workflow is something we though wouldn't give us this much problems. We spent too much time combining and integrating separate components and getting them all running simultaneously took longer than expected. Although we used Figma to plan, we learned that in order to complete on the schedule we set, we have to completed the entire workflow diagrams before starting and to finalize the conceptual design before implementation.

What's next for GemAds

Next we plan to ship this to as many people as possible to showcase the impact a chatbot which garners so much attention can be used for in the ad business. By allowing for targeted products to reach users, we can provide the best services to users, while also helping to provide an alternative revenue source, possibly allowing everybody to have access to the best models.