AWS Builder Center
Building WatchWise: Using Generative AI to Decide Whether a YouTube Video Is Worth Watching

Building WatchWise: Using Generative AI to Decide Whether a YouTube Video Is Worth Watching

Many YouTube videos have attractive titles but do not always deliver useful content. In this article, I share how I built WatchWise, an AI-powered system that analyzes video transcripts, evaluates content quality, and helps users decide whether a video is worth their time.

Computer Science @ SRMIST | Building backend systems & AI-powered applications

Building WatchWise: Using Generative AI to Decide Whether a YouTube Video Is Worth Watching

Introduction

Every day, millions of people watch educational, technical, and informational videos on YouTube.
The challenge is that not every video delivers on its title or thumbnail. Some videos contain valuable information, while others are repetitive, misleading, or provide very little value despite being long.
As a student interested in AI and Machine Learning, I wanted to explore whether Generative AI could help users make better decisions before investing their time in a video.
This idea led me to build WatchWise, an AI-powered YouTube Video Quality Analyzer.



The Problem

When searching for information on YouTube, users often face several challenges:
  • Clickbait titles and thumbnails
  • Long videos with limited useful information
  • Difficulty identifying high-quality educational content
  • Lack of a quick way to evaluate whether a video is worth watching
Most tools focus on generating summaries, but summaries alone do not answer an important question:
Is this video actually worth my time?
WatchWise was designed to address this problem.



The Idea Behind WatchWise

The goal of WatchWise is simple:
  1. Retrieve a video's transcript
  1. Analyze the content using Generative AI
  1. Evaluate the educational value and quality
  1. Generate useful insights for viewers
Instead of only summarizing the content, WatchWise attempts to determine:
  • Whether the title matches the actual content
  • Whether the video provides meaningful information
  • Key takeaways from the video
  • Strengths and weaknesses of the content
  • An overall quality assessment
This helps users quickly decide whether to watch, skip, or save a video.



Technologies Used

The project combines several technologies:

Backend

  • Python
  • FastAPI

AI and Analysis

  • Gemini AI
  • Prompt Engineering

Data Sources

  • YouTube Data API
  • YouTube Transcript API

Database

  • SQLite

Browser Integration

  • Chrome Extension



How WatchWise Works

The workflow is straightforward:

Step 1: Get Video Information

The system retrieves metadata such as:
  • Video title
  • Channel information
  • Description

Step 2: Extract Transcript

The transcript is collected using the YouTube Transcript API.
The transcript provides the actual content spoken in the video.

Step 3: AI Analysis

The transcript is sent to Gemini AI with carefully designed prompts.
The model evaluates:
  • Content relevance
  • Information quality
  • Educational value
  • Consistency with the title
  • Key insights

Step 4: Generate Results

WatchWise produces:
  • A summary
  • Quality assessment
  • Key takeaways
  • Recommendations



Challenges Faced

Building WatchWise taught me several practical lessons.

Prompt Design Matters

Different prompts produced different results.
Small prompt changes often had a significant impact on the quality of the analysis.

Long Context Handling

Some videos contain very large transcripts.
Managing large amounts of text while maintaining useful analysis was an important challenge.

Reducing Unnecessary AI Calls

Generative AI requests can become expensive.
To address this, I implemented SQLite-based caching to avoid repeated analysis of the same content.

Security Considerations

Since the project processes external content, security became important.
I implemented:
  • Input validation
  • Prompt injection protection
  • HTML sanitization
  • Rate limiting
  • Request size restrictions



Key Lessons Learned

This project reinforced several important ideas about AI engineering.

AI Is More Than Models

Building an AI application involves much more than calling an API.
It requires:
  • Data collection
  • Backend development
  • Security
  • Cost management
  • User experience design

Evaluation Is Difficult

Determining whether content is "good" is subjective.
Designing meaningful evaluation criteria was one of the most challenging parts of the project.

User Problems Matter Most

The goal was never to use AI for the sake of using AI.
The goal was to solve a simple problem:
Help users decide whether a video is worth watching.



Future Improvements

There are several directions for future development:
  • Personalized recommendations
  • Better credibility analysis
  • Community-driven feedback signals
  • Multi-language support
  • Improved educational quality scoring



Conclusion

WatchWise started as an experiment to explore Generative AI and content analysis.
Along the way, it became a valuable learning experience in AI engineering, backend development, security, and system design.
The project reminded me that successful AI applications are not only about choosing a powerful model. They are about understanding a real problem and building a reliable solution around it.
As Generative AI continues to evolve, I believe tools that help users evaluate information quality will become increasingly important.



Project Links

Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article