The Entourage

markdown## Inspiration

Everyone tries to reach celebrities directly. That never works. Their DMs are filtered, their emails are guarded, and their notifications are a wall of noise. But every major public figure has a manager, a publicist, an agent, and an assistant — and those people have normal inboxes, active social accounts, and it is literally their job to surface important things to the person they work for.

We asked ourselves: how do talent agencies actually get meetings with hard to reach people? They don't cold DM the celebrity. They call their manager. We built a tool that automates exactly that.

What it does

The Entourage takes any celebrity name and automatically maps their professional inner circle using public data. It finds real named individuals — managers, publicists, agents, assistants, collaborators — and scores each one by how reachable they actually are as a human being. It then writes a personalised outreach message for each contact and sends it. The entire flow from typing a name to having real outreach in a real inbox takes under 60 seconds.

How we built it

We built the full stack in 8 hours using Next.js and Tailwind on the frontend, Supabase for the database and real time outreach logging, and OpenAI GPT-4o for two things: parsing unstructured Google search snippets into clean structured contact data, and generating three personalised message variants per contact — warm, direct, and bold. Resend handles live email delivery. SerpAPI powers the targeted searches. Everything deploys on Vercel.

The pipeline works in four steps. First we hit the Wikipedia API for career context. Then we fire targeted Google searches for each role — manager, publicist, agent, assistant. Then GPT-4o parses the raw snippets into real named individuals with roles and platforms. Finally each contact is scored 1 to 10 by reachability and sorted so you always hit the most accessible person first.

Challenges we ran into

Getting GPT-4o to return consistent structured JSON from messy unstructured search snippets was harder than expected. We went through three prompt iterations before the output was reliable enough to use. Supabase Realtime required careful subscription management to update the outreach log instantly without flickering or duplicate rows. We also had to tune the scraping aggressively to avoid hallucinated contact names that sounded plausible but didn't exist.

Accomplishments that we're proud of

We built a fully working outreach pipeline in 8 hours that sends real emails to real people in real celebrity inner circles. We didn't just demo a prototype — we actually used the product live during the hackathon on real targets from the celebrity list. We made real calls. The product works beyond the demo environment and that matters to us.

What we learned

The shortest path to anyone is never direct. Reachability is inversely correlated with fame. The most useful person in a celebrity's circle is usually the least famous one. A well-researched 80 word message sent to the right person outperforms any amount of social media noise. And building a real end to end product in 8 hours is entirely possible if you cut scope ruthlessly, ship early, and iterate fast.

What's next for The Entourage

The use case extends far beyond a hackathon stunt. Sales teams, journalists, recruiters, and talent agencies all face the same problem — reaching high value hard to contact people efficiently. The next steps are building a Chrome extension that runs the pipeline on any LinkedIn or Twitter profile, adding response tracking so you can see when contacts open or reply, and expanding the outreach channels to include LinkedIn automated connection requests and WhatsApp where numbers are findable. The deeper vision is an AI powered relationship engine that doesn't just find the door — it learns which doors open fastest.

Built With

• claude
• css3
• javascript
• json
• lovable
• supabase
• tsx