⌂ HOME← ALL PROJECTS

Google Forms Scam Analyzer

Gemini APIGroqTavilyVirusTotal

Most security filters trust Google's domain, thousands of malicious phishing forms bypass automated checks.

Architecture & BYOK Security: Built a Manifest V3 Chrome extension and FastAPI backend where all LLM/OSINT API keys are supplied per-user (never stored server-side), transmitted over TLS as request headers, with per-IP rate limiting protecting shared demo keys from abuse. Multi-Provider LLM Pipeline: Implemented structured-output classification (Groq gpt-oss-120b primary, Gemini 3.6 Flash fallback) for scam-tactic detection and named-entity extraction, with provider order determined by live reliability testing rather than initial assumptions. Concurrent OSINT Integration: Engineered an async background-job pipeline (FastAPI BackgroundTasks) running four reputation APIs (Safe Browsing, VirusTotal, urlscan.io, Tavily) in parallel via ThreadPoolExecutor, decoupled from the synchronous request path. False-Positive Remediation: Diagnosed and fixed coincidental keyword-match false positives in entity reputation checks by redesigning the relevance judgement to require quoted, sourced evidence before flagging any entity. Deployed live on Render with a Postgres-ready SQLAlchemy data layer and content-hash caching to avoid redundant API calls.

A published extension on Chrome Web Store.