🌌 NovaStream

The Infinite AI Broadcast — 24/7, No Humans Required

NovaStream is a fully autonomous, AI-powered television network.
It continuously monitors the internet for breaking news, transforms headlines into
cinematic short-form video episodes — complete with scripted scenes, voiceover narration,
and matched stock footage — then broadcasts them live, 24/7, with zero human intervention.

---

✨ What is NovaStream?

NovaStream is an end-to-end agentic media pipeline built on Amazon Nova foundation models via AWS Bedrock. Give it the internet, and it gives you a television channel.

Every few minutes, the system:

1. 📰 Fetches breaking news headlines from the live web
2. 🎬 Produces a fully scripted, multi-scene episode using Amazon Nova 2 Lite
3. 🎙️ Voices the narration scene-by-scene using Amazon Nova 2 Sonic (TTS)
4. 🎞️ Casts matching cinematic stock footage via the Pexels API
5. ✂️ Edits everything into a final broadcast-ready video with FFmpeg
6. 📡 Broadcasts the episode live to a web player over WebSockets

---

🤖 Amazon Nova — The Intelligence Behind NovaStream

NovaStream is purpose-built around Amazon Nova, AWS's latest generation of frontier foundation models. Nova models are invoked through Amazon Bedrock — AWS's fully managed AI platform — and power every creative and generative step of the pipeline.

Nova Model	Model ID	Role in NovaStream
Amazon Nova 2 Lite	us.amazon.nova-2-lite-v1:0	The Showrunner — takes a raw news headline and produces a structured JSON production blueprint: scene count, scripts, visual themes, and narration copy
Amazon Nova 2 Sonic	us.amazon.nova-2-sonic-v1:0	The Voice Actor — performs hyper-realistic Text-to-Speech synthesis, narrating each scene with natural intonation and broadcast-quality audio

Why Amazon Nova?

• Speed — Nova Lite delivers sub-second structured JSON generation, keeping the pipeline latency low enough for continuous broadcasting.
• Quality — Nova Sonic produces broadcast-quality voice output indistinguishable from a professional narrator.
• Cost Efficiency — Nova's competitive token pricing makes running a 24/7 autonomous pipeline economically viable.
• Bedrock Integration — Seamless boto3 integration means no custom inference servers, no model hosting, and no ops overhead.

AWS Services Used

AWS Bedrock  ──►  Amazon Nova 2 Lite   (Showrunner: script & scene generation)
             ──►  Amazon Nova 2 Sonic  (Voice Actor: TTS narration synthesis)
AWS IAM      ──►  Secure credential management via boto3 / STS

---

🏗️ Architecture

graph TD
    A[📰 NewsAPI<br/>Breaking Headlines] --> B

    subgraph "🤖 Agent Pipeline (Python / FastAPI)"
        B[🎬 Showrunner Agent<br/>Amazon Nova 2 Lite] --> C
        C[🎭 Casting Director<br/>Pexels API + Ranking] --> D
        D[🎙️ Voice Actor Agent<br/>Amazon Nova 2 Sonic] --> E
        E[✂️ Editor Agent<br/>FFmpeg]
    end

    E --> F[☁️ Supabase Storage<br/>Episode CDN]
    E --> G[📡 WebSocket Broadcaster]
    G --> H[🖥️ Next.js Live Player<br/>localhost:3000]
    G --> H[🖥️ Next.js Live Player<br/>novaaastream.vercel.app]

Loading

Agent Breakdown

Agent	File	Responsibility
Showrunner	agents/showrunner.py	Nova 2 Lite prompt → structured JSON episode blueprint
Casting Director	agents/casting.py	Query expansion + Pexels video search + quality ranking
Voice Actor	agents/voice.py	Nova 2 Sonic streaming TTS → per-scene .mp3 audio
Editor	agents/editor.py	FFmpeg scene stitching → final .mp4 broadcast episode
Broadcaster	broadcaster.py	WebSocket push of episode metadata + video URL to frontend

---

💻 Tech Stack

Layer	Technology
AI Models	Amazon Nova 2 Lite, Amazon Nova 2 Sonic (via AWS Bedrock)
Backend	Python 3.11, FastAPI, Asyncio, WebSockets
Video Processing	FFmpeg 6.0+
Frontend	Next.js 14, React, Tailwind CSS
Storage	Supabase Storage
External APIs	NewsAPI, Pexels Video API
Infrastructure	Docker, docker-compose, Render

---

🚀 Quick Start

Prerequisites

• Python 3.11+
• Node.js 20+
• FFmpeg 6.0+ — brew install ffmpeg (macOS) / apt install ffmpeg (Linux)
• AWS credentials with us-east-1 Bedrock access (Nova 2 Lite + Nova 2 Sonic)
• Supabase project with a storage bucket
• Pexels API key
• NewsAPI key

1. Clone & Configure

git clone https://github.com/your-username/novastream.git
cd novastream
cp .env.example .env

Edit .env with your credentials:

# AWS (Bedrock — Nova models)
AWS_ACCESS_KEY_ID=...
AWS_SECRET_ACCESS_KEY=...
AWS_DEFAULT_REGION=us-east-1

# Nova model IDs
NOVA_LITE_MODEL_ID=us.amazon.nova-2-lite-v1:0
NOVA_SONIC_MODEL_ID=us.amazon.nova-2-sonic-v1:0

# External APIs
NEWSAPI_KEY=...
PEXELS_API_KEY=...

# Supabase Storage
SUPABASE_URL=...
SUPABASE_SERVICE_ROLE_KEY=...
SUPABASE_BUCKET=...

⚠️ .env is gitignored — your AWS keys will never be accidentally committed.

2. Run with Docker (Recommended)

docker-compose up --build

3. Run Manually

Backend:

cd backend
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
python main.py

Frontend (new terminal):

cd frontend
npm install
npm run dev

Open http://localhost:3000 — the infinite broadcast is live. 📺

---

⚙️ Configuration Reference

Variable	Description	Default
NOVA_MAX_SCENES_PER_EPISODE	Cap scenes per episode (cost control)	2
SHOWRUNNER_MAX_TOKENS	Max tokens for Nova Lite generation	500
SHOWRUNNER_TEMPERATURE	Nova Lite creativity (0.0–1.0)	0.4
SHOWRUNNER_MAX_RETRIES	Retry attempts on Bedrock failure	1

💡 Cost Tip: Keep NOVA_MAX_SCENES_PER_EPISODE=2 during development to minimize Bedrock token usage.

---

🏆 Hackathon

Built for the Amazon Nova AI Hackathon — March 2026.

• 🧠 Agentic AI — Fully autonomous multi-agent pipeline with no human checkpoints
• 🎙️ Multimodal Generation — Cross-model pipeline spanning text (Nova Lite), speech (Nova Sonic), and video (FFmpeg + Pexels)
• 📡 Real-Time Streaming — Live WebSocket broadcast of AI-generated television

---

📄 License

This project is licensed under the MIT License — see the LICENSE file for details.

---

_{Built with ❤️ and Amazon Nova · Powered by AWS Bedrock}