Hackathon Challenge: The "GEO" Agent (Live Data Edition) Jan 18th Sunday Hackathon 

Mission: Build the "SEO Tool" for the AI Era 

Time Limit: 6 Hours

For 20 years, brands lived and died by Google’s "10 blue links." That era is ending. Today, consumers ask ChatGPT, Perplexity, gemini, or Claude: "What is the best study abroad program or Coaching program for international students?" The AI generates a single, synthesized answer. If a brand isn’t mentioned there, it doesn’t exist.

The Problem: Brands have zero visibility. They don't know what AI models are saying about them, or if the information is even accurate.

The Client: The Director of the Global Edge Program (a study abroad & business education initiative).

The Pain Point: They have a great website, but when students ask ChatGPT "What are the best international business programs?", the AI recommends "Semester at Sea" and "NYU Florence," but completely ignores Global Edge, The Goal: The Director wants an agent that can:

1. Audit: Tell them exactly which AI models are ignoring them.

2. Explain: Reveal why (e.g., "Your competitor has a Wikipedia page; you do not").

3. Fix: Draft the content needed to get them into the AI's answer.

This is a "Live Fire" challenge.

• You cannot hard-code responses.

• You cannot use a pre-made JSON file of "fake search results."

• Your Agent must actually go online. It needs to make real API calls to live sources (Search tools, LLMs, or Web Scrapers) to fetch the current reality of the brand.

Build an AI Agent that audits, monitors, or optimizes a brand's "Share of Model" using real-time data.

Your agent acts as a consultant. It investigates how a brand appears across the internet right now and provides strategies to fix it.

Pick any real-world brand. Do not invent a fake company. You need a brand with a digital footprint so your agent can find real data.

• A Global Giant: (e.g., Nike, Coca-Cola) – Good for testing high-volume data.

• A Tech Startup: (e.g., Linear, Notion, Retool) – Good for testing technical accuracy.

• A Local/Niche Program: (e.g., Global Edge) – Good for testing "discoverability" problems.

Choose one "Agent Persona" to build. Remember: Real Inputs $\rightarrow$ Real Analysis.

Option A: The Auditor Agent 🕵️‍♂️

• Live Action: The Agent uses a search tool (like Tavily) to scrape the top 5 ranking articles for "Best Global Business Programs 2026," reads them, and checks if Global Edge is mentioned.

• Real Insight: "I scanned the top 5 sources Perplexity uses. Global Edge is mentioned in 0 of them. However, your competitor is mentioned in 3."

Option B: The "Vs" Agent 🥊

• Live Action: The Agent takes two URLs (globaledge.msu.edu vs competitor.com) and scrapes the text from both.

• Real Insight: "I compared your homepage to the competitor. They have a 'Curriculum' schema tag that the AI can read easily. You do not. That is why they rank higher."

Option C: The Fact-Checker ✅

• Live Action: The Agent queries a specific LLM (e.g., GPT-4) with 10 questions about the program.

• Real Insight: "I asked GPT-4 'Does Global Edge offer scholarships?'. It said 'No' (which is a hallucination). Here is the text snippet you need to add to your FAQ page to fix this."

To adhere to the "No Synthetic Data" rule, you will likely need one of these Retrieval Tools in your stack:

• Tavily API: Optimized for AI agents to search the web and extract clean text.

• Perplexity API: To get the "answer" currently being served to users.

• Exa.ai: For semantic search (finding similar links).

• Firecrawl: To turn any website URL into clean markdown for your agent to read.

1. Truthfulness: Did the agent pull real data from the web? (We will ask you to show the raw API response or source URL).

2. Latency vs. Depth: Did you balance speed with the depth of the search?

• Creativity
  How original and insightful is the approach?
• Commercializability
  Could a real company pay for this tomorrow?
• Technically Sound
  Is this technically sound and real?
• Presentation
  Did you communicate the value clearly?