GreenLight

• 

• 

• 

• 

• 

💡 Inspiration

Creators don't have an idea problem. They have a decision problem. A filmmaker, YouTuber, or small production team may have ten ideas but only enough time and budget to produce one. Yet that decision is often based on intuition, trends, or incomplete research. Most AI tools respond with:

“Great idea! Let me help you make it.” We wanted the opposite: an AI development executive that can say NO.

🎬 GREENLIGHT asks one question:

“Which idea actually deserves to be made — and why?” To make that decision intelligently, the agent needs more than model memory. It needs the live web. 🌐 That's why Parallel is the eyes of GREENLIGHT.

🎬 What it does

GREENLIGHT is an agentic pre-production intelligence system for creators and media teams. A user submits a media concept, and GREENLIGHT runs it through a virtual development room:

💡 Idea
↓
🎬 Producer — understands the concept and format
↓
🌐 Researcher — investigates the live web with Parallel
↓
🧠 Analyst — Gemini evaluates the evidence
↓
⚔️ Contrarian — actively tries to kill the idea
↓
📊 Scoring Engine — measures its opportunity
↓
🎥 Development Executive — makes the call

The result: 🟢 GREENLIGHT (80–100) — worth producing
🟡 REWORK (60–79) — opportunity exists, but change the angle
🔴 PASS (0–59) — don't spend production resources on this version

GREENLIGHT scores seven dimensions: Audience · Novelty · Momentum · Competition · Evidence · Feasibility · Differentiation The final score is calculated from those dimensions in code. It also provides: ⚔️ Contrarian Case — the strongest reason the idea could fail
💡 Better Angle — a stronger alternative concept
🎬 Production Blueprint — hook, narrative, scenes, B-roll, verification needs, and CTA
📎 Live Evidence — clickable sources retrieved through Parallel

The memorable difference:

🧨 Most AI helps you make an idea. GREENLIGHT decides whether you should.

🛠️ How we built it

GREENLIGHT is a focused multi-step agent workflow, not a chatbot.

💡 User Concept
      ↓
🎬 Producer
      ↓
🔎 Research Planning
      ↓
🌐 Parallel Search API
      ↓
🧠 Gemini Analysis
      ↓
⚔️ Contrarian Evaluation
      ↓
📊 Deterministic Scoring
      ↓
🟢 GREENLIGHT / 🟡 REWORK / 🔴 PASS
      ↓
🎬 Production Blueprint

🌐 Parallel — Live Web Intelligence

GREENLIGHT actively uses the Parallel Search API at runtime. The server calls: POST https://api.parallel.ai/v1beta/search Parallel results are transformed into: 📄 Source title
🌍 Domain
📝 Relevant excerpt
🔗 Original URL These become evidence for GREENLIGHT's analysis and are displayed directly in the verdict.

Parallel isn't decoration — it gives the agent current, traceable information from the open web. API keys remain server-side and are never exposed in the browser.

🧠 Gemini — Reasoning

Gemini reasons over the concept and research evidence to perform:

• Opportunity analysis
• Competition and differentiation analysis
• Contrarian evaluation
• Structured scoring
• Executive recommendation
• Production planning

💻 Stack

⚛️ React + TypeScript
🎨 Tailwind CSS
⚡ Vite
🧠 Gemini
🌐 Parallel Search API
▲ Vercel
🐙 GitHub

🧩 Challenges we ran into

🔐 Development vs. production: Parallel worked inside Google AI Studio but initially failed after Vercel deployment because the production application needed its own server-side API routes and environment secrets. We built Vercel serverless endpoints so the production flow became: Browser → Serverless API → Parallel → Gemini → Verdict 🖥️ Blank verdict: An empty successful API response caused the verdict screen to render nothing. We hardened the response handling so the application always produces a usable state. 🖱️ Design vs. usability: Some early 3D effects interfered with evidence links. We preserved the cinematic interface while ensuring every source remains clickable. ⏱️ Scope: Instead of building authentication, payments, databases, and dozens of agents, we focused on one complete workflow:

Idea → Research → Challenge → Evidence → Decision → Production Plan

🏆 Accomplishments that we're proud of

🌐 Real Parallel Search running in the deployed application 📎 Live, clickable evidence instead of fabricated citations 🧠 A multi-stage reasoning workflow rather than a simple prompt wrapper 🧨 An AI intentionally designed to disagree with the user 📊 Transparent scoring across seven dimensions 🎬 A complete path from idea evaluation to production blueprint* 🎨 **A cinematic “digital development room” rather than another chatbot UI 🚀 And most importantly — it works live.

📚 What we learned

👁️ Agents need eyes. Gemini can reason, but Parallel gives GREENLIGHT access to current external evidence. 🔗 Evidence builds trust. A recommendation becomes much stronger when users can inspect the sources behind it. ⚔️ AI doesn't always need to create more. Sometimes its highest-value decision is telling the user not to make something. 🧮 Determinism matters. The final GREENLIGHT score is derived from visible scoring dimensions instead of being an unexplained number. 🎯 Scope is a feature. One polished, working agent workflow is more valuable than ten unfinished features.

🚀 What's next for GreenLight

🥇 Idea tournaments — research and rank multiple concepts simultaneously 🔬 Deeper Parallel intelligence — use additional research/extraction workflows for richer evidence. 📈 Decision history — compare GREENLIGHT predictions with how published content actually performs. 👥 Collaborative development rooms — enable production teams to evaluate projects together. 🎥 Production integrations — transform approved concepts into research briefs, interview plans, shot lists, and production workflows.

The philosophy will remain simple:

🎬 Most AI asks: “What should I create for you?” 🟢 GREENLIGHT asks: “Should this be created at all?”

Before you produce it, let AI challenge it.

🌐 Live App: https://green-light1.vercel.app
🐙 Source Code: https://github.com/Iqra-Khan17/GreenLight

Built With

• agentic
• cenima
• gemini
• github
• hackathon
• parallel
• react
• tailwind-css
• typescript
• vercel
• vite