AdLumen

Inspiration 🌟

There has been an increasing trend of AI being used in advertisement and even scams and fraudulent businesses all across the world, and AI is only going to get better. The concerning lack of pro-consumer regulations drove us to develop a user friendly app to fight fire with fire. So we made AdLumen to detect whether your site is chill or not.

What it does 🔎

Using a single agent workflow with Human-In-the-Loop, AdLumen takes a URL and scans the credibility of the site based on the site's images, domain name and content, then gives the user an assessment on whether the site contains deceptive or fraudulent content or not.

How we built it 🛠️

We built the frontend using NextJS and a FastAPI backend to serve us the data, and built tools and orchestrated workflows for our agent in the backend.

Challenges we ran into ⚠️

We initially planned to get YouTube ad videos and do deepfake analyses of them using a Chrome extension. We pivoted from our ambitious plan to do site URLs instead.

What we achieved 🏆

• Oliver and Laura learnt backend API routing
• Henrique learnt NextJS, TailwindCSS
• Thomson learnt about Manifest V3

What we learned 📚

• TailWindCSS
• NextJS
• FastAPI
• Agents and how they work
• Manifest V3, and how chrome extensions work
• DOM query selectors and DOM manipulation

What's next for AdLumen 🚀

• Storing scan results with MongoDB, and connecting MongoMCP for our agent
• Connecting a background worker to automate getting URLs instead of manual inputs
• Taking a different approach to detect deceptive AI ads on YouTube instead of using a Chrome extension; perhaps getting ad transcripts and taking snapshots of the video at certain frames

Built With

• bson
• fastapi
• gemini
• mongodb
• nextjs
• python
• tailwindcss
• typescript