AI Labs

Contact Me
Contact Us

FLUFF CUTTER

Fluff Clutter

I built simple web tool that removes filler and fluff from YouTube videos so users can quickly get the core information.

The focus was speed, simplicity, and a single-page experience that anyone could use without installing software.

 

Key design decision

Instead of calling the YouTube Data API, the app retrieves transcripts directly using a transcript-reading library. This choice simplified setup and avoided API keys, quota limits, and additional billing. It also reduced friction for deployment and embedding.

 

The pipeline became:
YouTube URL → transcript reader → AI processing → condensed output.

 

Why transcript reading instead of the YouTube API

Reasons for this choice

  • No Google API setup or OAuth required

  • No quota or rate-limit concerns

  • Faster implementation and fewer moving parts

  • Simpler deployment to Cloud Run

  • Better fit for a lightweight public web tool

 

This decision kept the architecture lean and focused on the main goal: transforming transcripts into useful summaries.

Resulting AI workflow

Why this approach works well

This architecture balances:

  • Low complexity

  • Low cost

  • High reliability

  • Fast response time

  • Easy deployment and embedding

 

The result is a focused AI utility that does one job well: turning long YouTube videos into concise, useful content.

Key Features of the Approach:

By using Google AI Studios I was able to create an APP less than 20 minutes. The code was generated using AI and then fine tuned with AI input. 🚀

01.

Paste-and-go workflow
Users only need to paste a YouTube link. No accounts, no setup, no extra steps.

Transcript-first architecture
The app reads captions directly instead of using the YouTube API. This removes API keys, quotas, and setup friction while keeping the pipeline simple and reliable.

AI fluff removal
The model is guided to remove:

  • Intros and outros

  • Sponsorship segments

  • Rambling and repetition

  • Filler language and small talk

The output focuses on the actual information.

Few-shot prompting for consistent results
Examples inside the prompt teach the model exactly how to format and structure summaries. This produces predictable, clean output every time.

Fast cloud deployment
Hosted on Google Cloud Run for:

  • Automatic scaling

  • High availability

  • Quick response time

  • No server maintenance

Embeddable web app
The tool can run inside any website using an iframe, making it easy to use as a portfolio demo, lead magnet, or internal productivity tool.

Single-purpose simplicity
Designed to do one job well: convert long YouTube videos into concise, readable content.

 
 
Scroll to Top