I built an AI automated video generation agent using AWS. Here is how?

Project Overview – Wisdom Wealth Automated Video Generation System

This project is a fully automated, AI-powered video generation and publishing pipeline built on AWS. It generates finance-domain videos using generative AI, produces multimedia content, and automatically uploads them to YouTube.

Codebase Structure

The system consists of four independent repositories hosted on GitHub:

1. wisdom_wealth

2. youtube_video_upload_lambda

3. wisdom_wealth_cleaner

4. ec2_start_stop_lambda

Each repository is deployed independently through CI/CD.

CI/CD Pipeline

All repositories are integrated with AWS CodePipeline, which handles automated build and deployment.

Deployment targets:

• wisdom_wealth → Deployed to Amazon EC2

• youtube_video_upload_lambda → Deployed to AWS Lambda

• wisdom_wealth_cleaner → Deployed to AWS Lambda

• ec2_start_stop_lambda → Deployed to AWS Lambda

This separation ensures modularity, scalability, and independent lifecycle management.

Scheduled Automation Layer

A scheduled rule in Amazon EventBridge periodically triggers the ec2_start_stop_lambda.

This Lambda function:

• Starts the EC2 instance (if stopped)

• Initiates the video generation workflow inside the wisdom_wealth application

This ensures compute resources are only active when needed, optimizing cost.

AI-Powered Video Generation Workflow

Once triggered, the wisdom_wealth The application executes the following pipeline:

Script Generation

A finance-domain script is generated using a large language model from Amazon Bedrock.

Script Processing

The generated script is:

• Split into individual sentences

• Each sentence is processed independently

Multimedia Generation

For each sentence:

• Image generation via Amazon Bedrock image models

• Audio narration generated using Amazon Polly

This creates synchronized visual and audio assets for every sentence.

Video Stitching

All generated:

• Images

• Audio clips

are stitched together using a video editing engine running on EC2 to produce a final video file.

Storage and Metadata Management

After video generation:

• The final video is uploaded to Amazon S3

• Metadata (title, description, tags, status) is stored in Amazon DynamoDB

This metadata acts as the control layer for publishing and tracking upload status.

YouTube Publishing Pipeline

The youtube_video_upload_lambda runs periodically and:

1. Reads pending records from DynamoDB

2. Uploads the corresponding video from S3 to YouTube

3. Updates the upload status in DynamoDB

Videos are published on the channel:
https://www.youtube.com/@trillionaireequation2074

Failure Handling & Cleanup

If a YouTube upload fails:

• The wisdom_wealth_cleaner Lambda runs periodically

• It detects failed or stale records

• Performs cleanup and status correction in DynamoDB

This ensures system consistency and prevents stuck records.

Architectural Characteristics

• Fully automated

• Event-driven

• Serverless + EC2 hybrid architecture

• AI-powered content generation

• Cost-optimized through scheduled compute

• Modular repository structure

• Production-grade CI/CD