Auto_Post Agent

Turns your long videos into unstoppable viral machines — while you sleep.

Comment

Inspiration

What it doesWe create content on the side for YouTube and other platforms, but producing short-form content (both text and video) is time-consuming and inconsistent. We wanted to automate the entire process — from extracting viral moments to posting engaging content across platforms — so creators like us can focus on making, not managing.

This isn’t just our problem — millions of content creators struggle with the same challenge. That’s why we decided to solve it.

What We Learned How to integrate multiple APIs (YouTube, X, LinkedIn, and soon Threads) into a single workflow.

Techniques for detecting “hot moments” in videos using retention analytics, transcript sentiment, and audio excitement.

The art of A/B testing for virality — how small changes in hook, caption, or thumbnail can drive big differences in engagement.

Coordinating work in parallel using Git branching so multiple features (AutoClip and personalization) could be developed independently.

How We Built It ChatGPT 5o for API call through WhisperAI

Authentication & Social Integration

LinkedIn and X login with OAuth2, Threads integration in progress.

Secure token storage for safe posting.

Text Post Generation (Personalization Branch)

Scrapes tone, style, and personality from a creator’s existing posts.

Generates platform-specific text posts from long-form YouTube transcripts.

AutoClip (Viral Detection Branch)

Pulls retention and engagement data from YouTube API.

Detects hot moments via sentiment analysis + audience spike zones.

Generates 15–60 second shorts with branded captions and styling.

A/B Testing Engine

Creates multiple variants (different captions, hooks, thumbnails, post times).

Tracks performance and recommends the winning variant.

Deployment

Hosted backend handling the agent pipeline.

GitHub for version control, separate feature branches for parallel dev.

Challenges We Faced Data Access: YouTube API quotas limited how often we could pull analytics.

Variant Generation: Balancing speed vs. creativity when making multiple clip versions.

Multi-platform Posting: Each platform’s API had different rules and limits.

Built With

  • a/b
  • engagement
  • engine
  • ffmpeg-cloud-&-deployment:-brev-(nvidia-gpu-hosting)-github-for-version-control-databases-&-storage:-sqlite-(session-&-post-data)-cloud-storage-for-generated-clips-apis:-youtube-data-api-(analytics
  • for
  • javascript-frameworks-&-libraries:-backend:-fastapi
  • langchain-ai/ml:-openai-api-(gpt-4)
  • languages:-python
  • personalization)
  • sentiment-analysis-with-transformers
  • testing
  • transcript
  • video-retrieval)-x-(twitter)-api-linkedin-api-threads-api-(in-progress)-other-tools:-oauth2-for-authentication-git-branching-for-feature-isolation-(autoclip
  • whisper-api-for-transcription-video-processing:-moviepy
Share this project:

Updates