Building an Autonomous Media Intelligence System with Python

#python #ai #automation

Building an Autonomous Media Intelligence System

Overview

I've been building Nexus Intelligence — an autonomous system that discovers viral topics, finds contextual media, and produces YouTube Shorts automatically.

Architecture

The system has 3 components:

Media Detective — Scans Google Trends, Reddit, HN, RSS feeds for viral topics
Contextual Media Search — Uses Pexels API + Pixabay API to find licensed images/videos
Video Pipeline — Script generation → TTS → footage assembly → FFmpeg → YouTube upload

Key Features

Multi-Source Media Discovery

def search_contextual_media(self, query: str, count: int = 5):
    # Pexels (licensed, HD/4K)
    self.search_pexels_images(query, count)
    self.search_pexels_videos(query, count=3)
    # Pixabay (no attribution required)
    self.search_pixabay_images(query, count)
    self.search_pixabay_videos(query, count=3)
    # Bing fallback
    if len(all_discoveries) < count:
        self.search_bing_images(query, count)

Pre-Publish Quality Check

Before any video is uploaded:

Script sanitizer filters sensitive words
Voice-to-text decoder (Whisper) verifies the generated audio matches the script
LLM analyzes the content for quality and coherence

Evolutionary Trading Strategies

The system also includes an evolutionary trader that uses genetic algorithms to discover trading strategies:

Tournament selection, crossover, mutation, elitism
Walk-forward validation
Continual learning on real-time market data (every 5 minutes)
Multi-exchange data: Binance, Bybit, Kraken, Hyperliquid

Tech Stack

Python 3.11 — Core language
Ollama — Local LLM for script generation (qwen2.5:3b+)
Edge-TTS — Free Microsoft Neural voices
Pexels/Pixabay API — Licensed stock media
FFmpeg — Video assembly
YouTube Data API v3 — Upload
Whisper — Voice-to-text verification

Results

55.8% win rate on multi-coin backtest (BTC 67.6%)
First realized gains: +$0.0162 USDC
Automated media discovery across 5+ sources

Next Steps

Improve win rate to 80% for leveraged trading
Scale to 11 shorts/day
Add reverse image search for source verification
Add journalist-grade source feeds (AP, Reuters, Getty)

This is a project from Nexus Intelligence — an autonomous media and trading system.

DEV Community