FORGE — Art of original
FORGE hunts fake clone stores of small D2C brands, proves they're fake, and sends a signed evidence PDF and abuse report.
Built at AgentX Hackathon 2026
Team Shared21: Prarthana S., Surabhi Muddan
Built with Python, LangGraph, Gemini, Playwright, Streamlit, urlscan.io, dnstwist, crt.sh, RDAP, imagehash, ReportLab, pyHanko, NetworkX, SQLite, Resend
About the project
Problem: Small D2C brands lose sales and trust to clone stores that copy their product photos, logo and name, run Instagram ads, and vanish and relaunch on new domains. Brands rarely find them, and takedown requests get ignored without proof.
Solution: FORGE is an AI agent built on LangGraph. Give it a brand's store URL and it fingerprints every logo and product photo (SHA-256 + perceptual hash). Then it hunts for copies using image-hash and title search on urlscan, plus lookalike domains from dnstwist and crt.sh.
How it works: Each suspect is visited three ways: as a bot, as an Instagram-ad shopper, and from another country, to catch cloaking. Gemini reads the page. A scorer weighs copied photos, copied logo, prices far below the brand's, unofficial contacts, domain age and relaunches. It labels each store likely fake, suspicious or ignored. The owner confirms in a Streamlit dashboard. FORGE then sends a digitally signed evidence PDF and an abuse report, and learns from the owner's answers.
Impact: Proof a host or registrar can act on, in minutes instead of weeks.