🚀 AI-Powered Creator Intelligence System

Reverse-engineering long-form content performance using Python, YouTube Data API, Claude AI, and Power BI — across 545+ videos from one of India's largest podcast ecosystems.

---

📌 Project Overview

I analyzed 545+ Raj Shamani videos to understand what actually drives long-form content performance.

This project combines:

• Python automation
• YouTube Data API v3 extraction
• AI-assisted classification (Claude AI)
• Power BI intelligence dashboards
• Strategic recommendation modeling

The goal was not to build another analytics dashboard.

The goal was to reverse-engineer:

• virality mechanics
• audience psychology
• emotional trigger patterns
• hook structures
• title engineering
• and high-performing content formulas

The result is a complete Creator Intelligence System capable of surfacing strategic insights from one of India's largest long-form podcast ecosystems.

---

🖼️ Dashboard Preview

Page 1 — Creator Intelligence System

---

Page 2 — Strategic Recommendation Engine

---

📊 Channel Intelligence Captured

Metric	Value
Videos Analyzed	545+
Total Views	758M+
Average Views per Video	1.39M
Highest Viewed Video	36M views
Average Video Duration	76.5 Minutes
Analysis Timeline	2019–2026

---

🏗️ System Architecture

YouTube Data API v3
        │
        ▼
youtube_extractor.py          ← Fetches all videos, filters < 10 min
        │
        ▼
raj_shamani_longform_videos.xlsx  ← Raw dataset (545 videos × 9 columns)
        │
        ▼
Claude AI (Batch Classification)  ← Classifies titles across 8 dimensions
        │
        ▼
raj_shamani_ai_classified.xlsx    ← Enriched dataset (545 videos × 17 columns)
        │
        ▼
RSMIS.pbix                        ← Power BI Intelligence Dashboard
        │
        ▼
Strategic Recommendation Engine

---

📁 Repository Structure

RSMIS/
├── 📄 youtube_extractor.py               # Data pipeline: YouTube API → Excel
├── 📊 raj_shamani_longform_videos.xlsx   # Raw extracted video data
├── 🧠 raj_shamani_ai_classified.xlsx     # AI-enriched classified dataset
├── 📈 RSMIS.pbix                         # Power BI dashboard
└── 📘 README.md

---

⚙️ How the System Works

Step 1 — YouTube Data Extraction

youtube_extractor.py uses the YouTube Data API v3 to:

1. Search for the Raj Shamani channel and retrieve its channel_id
2. Fetch the channel's full uploads playlist
3. Loop through all videos and extract snippet, statistics, and contentDetails
4. Parse ISO 8601 duration strings (PT2H1M35S) into total minutes
5. Filter out videos under 10 minutes (Shorts, clips, promos)
6. Export structured data to Excel

Extracted Fields:

Column	Description
video_id	YouTube video ID
title	Full video title
published_at	Upload timestamp (ISO 8601)
views	Total view count
likes	Total like count
comments	Total comment count
duration	Raw ISO 8601 duration string
duration_minutes	Duration parsed into minutes
video_url	Full YouTube URL

---

Step 2 — AI-Assisted Classification

Every video title was passed through Claude AI for zero-shot classification across 8 strategic dimensions — no manual labeling required.

Dimension	What It Captures	Example Values
content_type	Broad content category	Education, Finance, Technology, Healthcare
main_topic	Specific subject	Indian Economy, Leadership, Engineering Careers
emotional_trigger	Primary emotion activated	Curiosity, Fear, Aspiration, Authority
hook_type	Opening hook mechanism	Curiosity Gap, Warning
title_style	Title structural pattern	Statement, Number-based, Interview, How-to, Question
audience_intent	Viewer motivation	Financial Growth, Learn, Entertainment
virality_potential	Predicted shareability	High, Medium, Low
guest_industry	Guest's professional background	Economist, Doctor, Sports, Educator

---

Step 3 — Power BI Intelligence System

The dashboard (RSMIS.pbix) transforms raw YouTube metadata into a multi-layer intelligence system with filters for date range, content type, and virality potential.

---

📊 Intelligence Layers Built

🎯 Attention Psychology Engine

Maps which emotional triggers generate the highest audience attention and view concentration.

🪝 Hook Structure Intelligence

Tracks how curiosity-gap framing outperforms warning-based hooks across the entire video catalog.

🧠 Topic Domination Mapping

Visualizes ecosystem-wide attention across AI, finance, education, and geopolitics.

📈 Virality vs Title Engineering

Identifies how title structure and word count influence content performance.

👥 Audience Intent Intelligence

Maps why audiences consume long-form content: learning, entertainment, financial growth, self-improvement.

🤖 Strategic Recommendation Engine

Generates content optimization insights based on discovered performance patterns.

---

🔍 Key Strategic Findings

Content Mix

The channel spans 10 content categories. Education and Technology dominate:

Content Type	Videos
Education	182
Technology	149
Finance	63
Business	38
Entertainment	34
Self Improvement	33
Healthcare	25
Psychology	16

---

Hook Strategy

• 91.6% of titles use a Curiosity Gap hook
• 8.4% use a Warning hook
• No videos rely on clickbait questions or unsubstantiated hype

---

Emotional Triggers

Trigger	Videos	Share
Curiosity	428	78.5%
Fear	46	8.4%
Authority	46	8.4%
Aspiration	25	4.6%

---

Title Formats

Format	Videos	Share
Statement	207	38%
Number-based	186	34%
Interview	75	14%
How-to	44	8%
Question	33	6%

---

Audience Intent

Intent	Videos
Financial Growth	212
Learn	199
Startup Learning	38
Self Improvement	37

---

Top Performing Videos

Views	Title
36M	Khan Sir Podcast: India vs China, Pakistan, Bihar's Reality & Geopolitics
30.4M	Vijay Mallya Podcast: Rise & Downfall Of Kingfisher Airlines
27.2M	Indian Spy: Dark Reality China, Weapons, Commando Training & Jail
14.3M	Dr. Cuterus on Sexual Health, Orgasm, G-Spot & Infertility
13.7M	Khan Sir on World War 3, India vs Pakistan, China, Trump & Epstein

---

Title Engineering

High-performing titles consistently clustered within the 15–18 word range. Statement and number-based formats dominated the top-performing tier.

---

🧠 Ideal Viral Podcast Formula

Element	Recommendation
Topic	AI / Finance / Geopolitics
Emotional Trigger	Fear + Curiosity
Hook Structure	Curiosity Gap
Title Length	15–18 Words
Title Format	Statement or Number-based
Audience Intent	Learn / Financial Growth
Format Style	Interview

---

🚀 Getting Started

Prerequisites

pip install google-api-python-client pandas openpyxl

Clone the Repository

git clone https://github.com/yourusername/RSMIS.git
cd RSMIS

Add YouTube API Key

Inside youtube_extractor.py:

API_KEY = "YOUR_API_KEY_HERE"

💡 Get a free YouTube Data API v3 key from Google Cloud Console

Run Extraction Pipeline

python youtube_extractor.py

Open Dashboard

Open RSMIS.pbix in Power BI Desktop (free download from Microsoft).

---

🔧 Configuration

In youtube_extractor.py, you can modify:

# Minimum video length to include (default: 10 minutes)
if total_minutes < 10:
    continue

# Change target channel
q = "Raj Shamani"  # Replace with any creator name

---

📦 Dataset Schema

raj_shamani_ai_classified.xlsx — full enriched dataset with 17 columns × 545 rows.

All AI classification was performed zero-shot on video titles only. No manual labeling was required.

---

🛠️ Tech Stack

Tool	Purpose
Python 3	Data extraction & automation
YouTube Data API v3	Video metadata source
google-api-python-client	YouTube API wrapper
pandas	Data manipulation & processing
openpyxl	Excel export
Claude AI	Zero-shot content classification
Power BI	Intelligence dashboard & visualization

---

💡 Future Improvements

• Thumbnail intelligence analysis (color, face detection, text overlay)
• View velocity tracking (views per day since upload)
• NLP on video titles (keyword frequency, bigrams)
• Comment sentiment analysis
• Competitor channel comparison (Nikhil Kamath, Ankur Warikoo)
• Automated weekly data refresh via GitHub Actions
• AI-generated content recommendations
• Retention pattern prediction

---

📄 License

MIT License. Use freely, attribution appreciated.

---

🙌 Acknowledgements

Built as a strategic content intelligence study on one of India's most consistent long-form YouTube creators. All data sourced via the official YouTube Data API v3.