CreatorMind

Know what to create before you create it.

Comment

Inspiration

YouTube creators face a persistent challenge: understanding what content resonates with their audience. While analytics dashboards provide raw numbers, they fail to answer the fundamental question every creator asks—"What should I create next?"

I observed creators spending hours manually reading comments, tracking trends, and guessing what their audience wants. This inefficiency inspired CreatorMind: a platform that transforms a creator's historical data into actionable intelligence, eliminating guesswork from content planning.

What it does

CreatorMind analyzes a creator's entire YouTube channel—every video, transcript, and comment—to generate data-driven content recommendations. The platform:

  • Connects to YouTube channels via OAuth and fetches complete video history
  • Extracts metadata (views, likes, comments, transcripts) and stores it permanently
  • Uses AI to analyze video characteristics: topics, tone, hooks, format, and complexity
  • Performs multilingual sentiment analysis and intent detection on all comments
  • Identifies performance patterns across formats, topics, tones, and upload timing
  • Generates weekly Top 5 video ideas with evidence-based reasoning and confidence scores
  • Delivers insights via interactive dashboard and automated weekly emails

Each video idea includes specific justification such as "42 viewers requested this topic" or "Similar videos achieved 2.1× your average engagement."

How I built it

Technology Stack

  • Frontend: React with modern UI components
  • Backend: Node.js with Express
  • Database: MongoDB for persistent storage
  • AI/ML: Google Gemini API for content analysis
  • APIs: YouTube Data API v3, OAuth 2.0
  • Email: React Email templates with automated scheduling

Architecture

The system follows a three-phase pipeline:

Phase 1: Data Collection

  1. User authorizes YouTube channel access via Google OAuth
  2. System fetches all videos with pagination (handles channels with n > 1000 videos)
  3. For each video v_i, collect: {title, description, views, likes, comments, transcript}
  4. Display real-time progress: "Processing i of n videos"

Phase 2: AI Analysis

  1. Extract video features using Gemini:

    • Topics and subtopics
    • Tone classification
    • Hook type identification
    • Format categorization
    • Complexity level (beginner | intermediate | advanced)

  2. Analyze comments across all videos:

    • Sentiment: S(c) ∈ {positive, neutral, negative}
    • Intent: I(c) ∈ {question, praise, request, criticism, confusion}
    • Topic extraction and clustering

Phase 3: Pattern Recognition & Idea Generation

  1. Calculate engagement rate: 

    E(v) = (likes + comments) / views

  2. Identify best-performing characteristics by computing:

    Performance(feature) = Σ E(v) / |V_feature|

    where the sum is over all videos with that feature.

  3. Generate ideas by matching:

    • Comment requests (demand signal)
    • Historical performance patterns (success indicators)
    • Audience profile (relevance filter)

  4. Assign confidence score: 

    C = w₁ · demand + w₂ · historical + w₃ · relevance

Challenges I ran into

API Rate Limits

YouTube Data API has quota limits (10,000 units/day). Fetching a channel with 500 videos and their comments could exhaust the quota. I implemented:

  • Intelligent pagination with exponential backoff
  • One-time full sync, then incremental updates
  • Prioritized data fetching (metadata first, transcripts second)

Transcript Extraction

Not all videos have auto-generated captions. I built a fallback system:

  1. Attempt YouTube's auto-caption API
  2. Fall back to yt-dlp extraction
  3. Mark videos without transcripts and analyze metadata only

Comment Clustering at Scale

Channels with 100,000+ comments posed computational challenges. I optimized by:

  • Batching comments for AI analysis (100 per batch)
  • Using embeddings for semantic clustering instead of pairwise comparison
  • Reducing complexity from O(n²) to O(n log n)

AI Hallucination Prevention

Gemini occasionally generated ideas without proper evidence. I mitigated this by:

  • Requiring explicit evidence linking (specific comment IDs, video titles)
  • Implementing verification layer: cross-check AI claims against database
  • Rejecting ideas with confidence < 40%

Email Scheduling Across Timezones

Users worldwide need emails at their local time. I implemented:

  • Timezone-aware scheduling with cron jobs
  • User preference storage (day, time, frequency)
  • Beautiful HTML emails using React Email templates

Accomplishments that I'm proud of

I successfully built a complete MVP that:

  • Processes unlimited videos from any YouTube channel
  • Performs sophisticated AI analysis on video content and audience feedback
  • Generates actionable, evidence-backed content ideas
  • Delivers insights through both web dashboard and automated emails
  • Handles edge cases (missing transcripts, rate limits, large channels)

The platform transforms the content creation workflow from reactive ("what did well?") to proactive ("what will work?").

What I learned

Technical Learnings

  • OAuth 2.0 implementation for secure third-party authorization
  • Working with large-scale data processing (handling 10,000+ videos × 100+ comments)
  • Prompt engineering for structured AI outputs
  • MongoDB aggregation pipelines for complex analytics
  • Background job processing and queue management

Product Insights

  • Users value why over what—showing evidence is crucial
  • Real-time progress indicators reduce perceived wait time during sync
  • Weekly cadence prevents information overload better than daily updates
  • Specificity wins: "42 viewers asked" is more compelling than "popular topic"

What's next for CreatorMind

Future enhancements include:

  • Competitor Analysis: Compare performance against similar channels
  • Thumbnail Intelligence: Analyze visual elements of high-performing thumbnails
  • Trend Detection: Identify emerging topics before they peak
  • Collaboration Suggestions: Recommend creators for potential collaborations
  • Multi-Platform Support: Extend to TikTok, Instagram, and podcasts
  • A/B Testing Framework: Test title/thumbnail variants with predicted outcomes

CreatorMind demonstrates that the future of content creation is not about guessing—it's about knowing. By combining comprehensive data analysis with AI-powered insights, creators can focus on what they do best: creating great content.

Built With

Share this project:

Updates