Participating in HackSeoul 2025 by Coupang & HackAngel was one of the most memorable experiences of my year. I joined the hackathon hoping to challenge myself, learn from others, and build an AI-powered, practical business model. I'm proud that our team was selected as a finalist by Coupang from more than 150 participant members.

---

🔍 Finding the Real-World Problem

We started the hackathon by searching for a real-world problem that AI could solve. Our team reviewed over 100 ideas, researched different industries, and discussed with mentors. Finally, we found a strong opportunity in the B2B marketing field:

measuring the real impact of offline advertising.

---

❗ The Problem: Offline Ads Can’t Prove Their Impact

Large companies spend huge budgets on street billboards. But they still can’t answer basic questions:

• Who actually saw the billboard?
• How long did they look at it?
• Did the viewer become a customer afterward?

There are no analytics tools like Google Analytics for offline advertising. So businesses have no way to measure ROI.

---

💡 Our Solution → AdVision AI

We proposed AdVision AI, an AI-powered offline advertising analytics solution. Our main features:

1. Attention Count – How many people made eye contact with the billboard
2. Focus Duration – How long each person looked
3. Conversion – How many of those people later became store customers This gives businesses clear insight into which billboard locations actually work.

🛠️ Building the Prototype

We had a 4-person team:

• Me → AI model integration, face detection, face embeddings, backend communication
• Kumush (Full Stack Developer / PM) → Web Dashboard & Real time Graph Visualization Integration
• Mark (Mobile Developer) → Camera + on-device detection
• Seo Hyun (Business Strategist) → Market Research

---

🧱 System Architecture

Below is the simple MVP architecture I built during the hackathon:

• AI Model Workers → Detect faces and generate face embeddings
• Backend (FastAPI + RabbitMQ + PostgreSQL) → Handles data flow between camera app, AI workers, and dashboard
• Camera Application (CCTV role) → Detects people, tracks eye contact, sends data
• Dashboard → Shows real-time analytics for each billboard

---

📊 Hackathon Result

After 24 hours with almost no sleep, we built:

• A real-time dashboard showing analytics for each billboard
• An AI model that collects face embeddings and focus duration
• A mobile camera app that detects people and sends data to backend

---

📝 Feedback From Judges

Strengths:

• The idea introduces a new way to measure billboard exposure that didn’t exist before.
• Strong potential in the B2B marketing market.

Considerations:

• Need to calculate maintenance cost vs existing CCTV competitors.

---

🎉 Conclusion

I’m proud that our team became one of the finalists at HackSeoul 2025. Everyone worked incredibly hard and stayed focused for the full 24 hours to complete the demo and present our project. The judges also found our idea innovative and meaningful for real-world business use. This experience gave me a lot of confidence and motivation to continue building AI-driven solutions that solve real problems.