This blog post is about my submission for Google GenAI Capstone Submission

➡️ Submission Public Notebook

Meet Your New Cyber Sidekick: The AI OSINT Recon Agent (No API Keys Were Harmed in the Making)

By xyizko — CyberN00b

"Why brute force when you can be... charmingly intelligent?"
– Probably James Bond if he was a threat analyst.

🎯 The Mission

Welcome, fellow cyber sleuths! Today, we’re unveiling a handy little tool born from the union of GenAI wizardry and some good ol’ fashioned Google Dorking — no sketchy scrapers or expensive APIs needed.

Say hello to the AI OSINT Recon Agent: a Kaggle notebook-powered digital detective 🧠🔍 that passively investigates domains, emails, or usernames with flair.

Whether you're a pentester prepping for recon, a bug bounty hunter scoping out targets, or just really nosy about your own digital footprint — this is your jam.

💡 What Does It Do?

Given a target (like example.com, admin@example.com, or even cyber_god99), this tool will:

1. 🤖 Classify what you gave it (email, domain, or username).

2. 🔎 Generate advanced Google Dork queries to dig up juicy info.

3. 🧾 Accept manually collected search snippets (we’re keeping it ethical!).

4. 📦 Extract structured intelligence (JSON!) from messy search text.

5. 🧙‍♂️ Generate a markdown report that's professional enough to submit to your CISO… or just hang on your wall.

🧬 GenAI Capabilities Used

Let’s geek out for a moment. We used 3 core capabilities from Google’s GenAI stack, all via Google AI Studio.:

1. 🧠 Few-Shot Prompting

We teach the model what’s what by showing examples — so it knows the difference between a domain and a username faster than you can say "input validation."

2. 🧾 Structured Output (JSON Mode)

We turn chaos (search snippet soup) into order (neat JSON lists of emails, leaks, GitHub links). Your SIEM will thank you.

3. 📚 Document Understanding

GoogleAI reads your pasted search snippets like a bored SOC analyst on espresso and tells you what matters — fast.

🧪 Real-World Applications

✅ Pros

• 🌍 No rate-limited APIs or shady scraping

• 🤝 Ethical and passive (great for red team pre-work)

• 🧠 Human-readable output + structured format

• 🧪 Easy to extend into a full agent workflow

⚠️ Cons (a.k.a. Room for Growth)

• 🤲 Manual snippet copy-paste — we stay legal, but it’s not 100% automated

• 🧱 Limited to the creativity of your Google Dorks (but hey, that’s an art!)

• 🤖 LLM hallucinations are rare but still possible — always verify before action

📦 What’s Next?

Glad you asked.

• Add a vector store + RAG to analyze hundreds of snippets

• Auto-pull Google results using a headless browser (ethically, of course 😇)

• Package it into a web interface for easier use by analysts

• Expand report output to PDF, email, or direct-to-Slack formats

🧁 TL;DR

This project blends classic recon techniques with modern GenAI brains to give you a low-lift, high-impact OSINT tool.

It’s like having an intern who never sleeps, never leaks data, and always returns JSON. 🚀

Want the code? Need help running it on your shady cousin’s domain?
Reach out, fork it, break it, or meme it. Just don’t ignore the power of passive recon.

Until next time, keep it passive. Keep it classy.
— The AI OSINT Recon Agent 🕶️