Efficient Video Scraping with Google Gemini: Extracting JSON Data for Under 1/10th of a Cent

Tools & Engineering

October 17, 2024

The other day, I faced a common but tedious task: adding up numeric values scattered across twelve different emails. Instead of manually copying and pasting each number, I decided to try something innovative: using screen capture and Google Gemini to extract the data.

AI Studio and QuickTime

I started by recording a 35-second screen capture using QuickTime Player on my Mac. The process was straightforward:

• Step 1: Open QuickTime Player.
• Step 2: Go to File -> New Screen Recording.
• Step 3: Drag a box around the portion of the screen containing my Gmail account.
• Step 4: Click through each email, pausing for a couple of seconds on each one.

Next, I uploaded the video to Google’s AI Studio, an AI-powered tool that can process and analyze various types of data. I prompted Gemini with the following request:

Turn this into a JSON array where each item has a yyyy-mm-dd date and a floating point dollar amount for that date

To my surprise, it worked flawlessly. The output was a well-structured JSON array:

[
  {
    "date": "2023-01-01",
    "amount": 2...
  },
  ...
]

Converting to CSV

Since I needed the data in a format that could be easily pasted into Numbers, I followed up with another prompt:

turn that into copy-pastable csv

Gemini delivered the same data formatted as a CSV:

date,amount
2023-01-01,2...
...

Accuracy Check

To ensure accuracy, I re-watched the 35-second video and manually verified the numbers. Gemini got everything right.

Cost Analysis

I intended to use Gemini 1.5 Pro, Google’s most advanced model, but accidentally used Gemini 1.5 Flash 002, a much less expensive option. According to AI Studio, I used 11,018 tokens, with 10,326 tokens dedicated to the video.

The cost breakdown:

• Tokens Used: 11,018
• Token Price (Gemini 1.5 Flash): $0.075 per million tokens
• Calculation:
  • ( \frac{11,018}{1,000,000} = 0.011018 )
  • ( 0.011018 \times 0.075 = $0.00082635 )

This means the entire process should have cost just under 1/10th of a cent.

Free Usage

However, it turned out to be free. Google AI Studio currently remains free of charge regardless of whether you set up billing across all supported regions. This is part of their effort to encourage users to try their tools, though it also means they can train on your data, which is not the case for their paid APIs.

Conclusion

This experiment demonstrates the power and efficiency of modern AI tools like Google Gemini in handling complex tasks with minimal effort and cost. Whether you're dealing with emails, documents, or other data sources, these tools can significantly streamline your workflow.