launch-posts/producthunt-assets.md

# Product Hunt Launch Assets

## OPTICPARSE — AI Vision Web Scraper

**Tagline:** Scrape any website with plain English. No CSS selectors. No XPath. Just ask.

**Description (200 words):**
Opticparse is an AI-powered web scraping API that eliminates the most painful part of web scraping: writing and maintaining CSS selectors and XPath expressions.

Instead, Opticparse uses Playwright to visit your target URL, captures a screenshot, and sends it to a vision AI model. You describe what data you want in plain English — the AI figures out the rest.

Works on JavaScript-heavy SPAs, handles redirects automatically, and won't break when a site changes its HTML structure.

**Key features:**
→ Natural language extraction queries
→ Handles dynamic JavaScript pages (React, Vue, Angular)  
→ Anti-bot stealth mode (bypasses basic Cloudflare detection)
→ Multi-provider AI (Groq, Vision AI (Groq LLaMA Vision + Gemini), fallbacks) — fast & reliable
→ 5-minute response caching for repeated requests
→ Available on opticparse.com with Free tier

**Best for:** developers who need to extract structured data from websites without the maintenance overhead of traditional scrapers.

**Website:** https://opticparse.com.com/parastejpal987cmyk/api/opticparse-ai-vision-web-scraper
**GitHub:** https://github.com/parastejpal987-cmyk/opticparse

---

**Topics:** Developer Tools, APIs, AI, Productivity

**First Comment (maker comment):**
Hey PH! 👋 

I built Opticparse after spending hours maintaining CSS selectors for a data pipeline that kept breaking every time the target site updated their frontend.

The insight was simple: if I can look at a screenshot of a website and know exactly where the data is, why can't an AI do the same thing?

The current version is fully production-ready on Render (free tier, kept warm 24/7 via uptime monitoring so no cold starts), uses Groq for sub-second AI inference, and has automatic fallback to Vision AI (Groq LLaMA Vision + Gemini) via GitHub Models and OpenRouter.

Free tier on opticparse.com — try it without a credit card.

What sites are you trying to scrape? Happy to help with queries in the comments!

---

## PHISHVISION — AI Phishing & Prompt Injection Detector

**Tagline:** Detect phishing pages and hidden AI prompt injection attacks using Vision AI (Groq LLaMA Vision + Gemini) vision

**Description (200 words):**
PhishVision is a cybersecurity API that catches what traditional phishing detectors miss.

Most phishing tools check URL reputation databases — useless for newly registered domains. PhishVision actually visits the URL with a headless browser, captures a screenshot, and analyzes it with Vision AI (Groq LLaMA Vision + Gemini) Vision using a forensic analyst prompt.

It also detects a newer, growing threat: **hidden AI prompt injection payloads** — text embedded in webpages (white-on-white, `display:none`, CSS-hidden) designed to hijack AI agents and chatbots that browse the web.

**Returns a structured JSON verdict:**
→ verdict: malicious / suspicious / safe  
→ confidence score (0–100%)
→ impersonated brand (if any)
→ threat type: brand_impersonation / prompt_injection / multiple / none
→ visual anomalies list
→ detected hidden payload text

**Key differentiators:**
→ Vision-based analysis (not URL reputation)
→ Detects prompt injection attacks specifically
→ Works on newly registered phishing domains
→ Free tier available via opticparse.com

**Website:** https://opticparse.com.com/parastejpal987cmyk/api/phishvision
**GitHub:** https://github.com/parastejpal987-cmyk/opticparse

---

**Topics:** Cybersecurity, AI, APIs, Developer Tools

**First Comment (maker comment):**
Hey Product Hunt! 

I built PhishVision after reading about a new attack vector that most security teams don't know about yet: **AI prompt injection via webpage content**.

Attackers are embedding hidden instructions like "IGNORE ALL PREVIOUS INSTRUCTIONS" in webpages using CSS tricks (white text on white background, `display:none`, 1px font-size). When an AI agent or chatbot visits the page, it can be hijacked.

URL reputation scanners (VirusTotal, PhishTank) don't catch this because they analyze the URL, not the content.

PhishVision actually renders the page with a real browser, extracts ALL text including hidden elements, and asks Vision AI (Groq LLaMA Vision + Gemini) to specifically look for these patterns.

The API is free to try — no credit card needed via opticparse.com.

Would love to hear from security teams about use cases I haven't thought of yet!