Undercut

💡 Inspiration Buying a used car is one of the most stressful financial decisions a person can make. Dealers have a massive data advantage, while everyday buyers are left guessing if a car is a "diamond in the rough" or a "money pit." We built Undercut to level the playing field and give every buyer the expertise of a professional car flipper.

🚀 What it does Undercut is your personal car-buying wingman. It proactively hunts for the best deals across major marketplaces using custom-built scrapers. Once a lead is found, Google Gemini AI performs a deep "Vibe Check"—analyzing descriptions for hidden mechanical red flags and comparing the deal against your specific, personal requests (like "Leather seats" or "Low insurance"). It doesn't just show you cars; it prepares you to win with AI-generated negotiation scripts tailored to that specific listing.

🛠️ How we built it We opted for a high-performance Hybrid Architecture:

The Hunter (Go): A high-concurrency scraper built in Go using Playwright to navigate dynamic marketplaces with stealth. The Controller (FastAPI): A Python backend that manages data flows and integrates the Google Gemini 3.0 Flash model. The UI (Next.js): A premium, glassmorphic frontend designed for speed and clarity. Persistence: A cloud-hosted Supabase (PostgreSQL) database. 🚧 Challenges we faced Our biggest hurdle was "Stealth Scraping." Marketplaces are designed to keep data locked away. We had to implement advanced anti-detection patterns and random jitter delays in our Go scraper to ensure reliable data flow without being flagged. Additionally, prompt-engineering Gemini to be "brutally honest" rather than generic required multiple iterations to get that expert "flipper" personality.

🏆 Accomplishments we're proud of We successfully built a full-stack, end-to-end pipeline that takes a raw, messy car listing and transforms it into a personalized, strategic negotiation script in under 5 seconds.

🧠 What we learned We learned the power of combining Go’s raw speed for data collection with Python’s AI ecosystem. Managing a monorepo across different runtimes (Go, Python, TypeScript) was a masterclass in project orchestration.

🔮 What's next for Undercut We plan to add real-time SMS alerts for "Instant Buys" and expand our AI to analyze vehicle history reports (VIN checks) to find even deeper hidden issues.