CekDulu

Inspiration

Online marketplace fraud is a real problem that hits close to many of us, or people we know, have likely experienced firsthand. Instead of leaving buyers to guess whether a listing is scam or not, we wanted to build something that answers that question instantly, before any money changes hands. The idea was simple: if technology can help people pause and think twice before checkout, why not build it?

What it does

CekDulu checks whether a product a buyer wants to purchase on a marketplace is potentially a scam or not. Users provide product information like item name, price, store, reviews and the system analyzes whether the listing shows signs of fraud from unreasonable pricing, to questionable seller credibility, to suspicious review patterns then delivers a clear verdict: safe to buy, or worth being cautious about.

How we built it

Frontend: Next.js + Tailwind CSS + shadcn/ui Backend: Django REST Framework + Celery + Redis Database: PostgreSQL via NeonDB Data Collection: Playwright Python AI Layer: LLM API (Deepseek V4 Flash)

Challenges we ran into

Our biggest challenge was finding the right data to work with. No public dataset provides direct scam vs. not scam labels for Indonesian marketplace listings, most available fraud datasets focus on credit card transactions or other data that isn't relevant to the local marketplace context. We had to search for and combine whatever data sources did exist like product listings and user reviews to build our own heuristics, instead of relying on a ready-made labeled dataset that turned out not to exist.

Accomplishments that we're proud of

We're proud that this project can help someone from being scammed, even in a small way, a real problem faced by everyday people, especially those who aren't very familiar with how marketplaces work, by helping them avoid falling victim to scams.

What we learned

We learned how to work as a team under time pressure, how to build a website from scratch, and maybe most importantly how to think through a real everyday problem and turn it into something functional, rather than just theoretical.

What's next for CekDulu

Next, we want to make the AI model more accurate and the product easier to use refining how it detects fraud signals and smoothing out the user experience so more people can actually rely on it before making a purchase.

Built With

• django
• neon
• openai
• typescript