𝗦𝘁𝗼𝗽 𝗙𝗶𝗴𝗵𝘁𝗶𝗻𝗴 𝗥𝗲𝗴𝗲𝘅 𝗪𝗶𝘁𝗵 𝗟𝗟𝗠𝘀

𝗦𝘁𝗼𝗽 𝗙𝗶𝗴𝗵𝘁𝗶𝗻𝗴 𝗥𝗲𝗴𝗲𝘅 𝗪𝗶𝘁𝗵 𝗟𝗟𝗠𝘀

I spent three days building a regex monster. It had 47 patterns. One missing space broke everything. I wanted to throw my laptop.

I tried a large language model. I used JSON mode. I told the model what fields I wanted.

I needed three things from emails:

• Order ID
• Intent
• SKU

Regex worked for 20% of emails. Real data is messy. Some orders had letters. Some SKUs were written differently.

I used gpt-4o-mini. It is fast. It is cheap. A few lines of code replaced my 47 patterns.

Follow these steps for reliability:

• Validate the JSON schema.
• Add 3 examples to your prompt.
• Log every failure.

LLMs are not for every task. Use regex for CSV files. Be aware of latency. Check your privacy rules.

Prompting is the new regex. It is easier to maintain. You change a prompt in seconds. Changing regex often breaks other things.

Start with the cheapest model. Mix regex for simple parts. Use LLMs for messy text.

Source: https://dev.to/__c1b9e06dc90a7e0a676b/how-i-stopped-fighting-regex-and-finally-extracted-data-with-llms-555i

Optional learning community: https://t.me/GyaanSetuAi