How do I determine if a text extraction needs an LLM?

This skill uses a confidence scorer that flags extractions based on missing fields, character length, or pattern mismatches, triggering an LLM call only when these thresholds aren't met.

Why use Regex instead of just using an LLM for all parsing?

Regex is significantly faster, cheaper, and 100% deterministic. Using it for the predictable 95% of your data reduces operational costs by over 95% while providing consistent results.

Which LLM model is best for the validation step?

Lightweight and fast models like Claude 3 Haiku are recommended, as they are highly capable of fixing specific extraction errors at a fraction of the cost of larger models.

Does this work for completely unstructured documents?

This framework is optimized for text with repeating patterns. For highly variable, free-form text, the skill recommends skipping regex and using an LLM directly.

Regex vs LLM Structured Text Parser

Name: Regex vs LLM Structured Text Parser
Author: affaan-m

byaffaan-m

•

112,914

•

ウェブスクレイピングとデータ収集

Optimizes structured text extraction by combining deterministic regex patterns with LLM validation for complex edge cases.

This skill provides a strategic decision framework and architectural pattern for parsing structured text like quizzes, forms, and invoices. By prioritizing high-speed regex for the 95-98% of cases that follow consistent patterns and reserving expensive LLM calls only for low-confidence edge cases, developers can achieve massive cost savings without sacrificing accuracy. It includes ready-to-use Python implementations for confidence scoring, hybrid pipelines, and standardized data models suitable for production-grade text processing.

主な機能

0195%+ cost reduction compared to LLM-only pipelines

02Hybrid parsing architecture combining Regex and LLMs

03Programmatic confidence scoring to identify extraction errors

04Detailed metrics for success rates and pipeline health

05112,914 GitHub stars

06Implementation patterns for quizzes, forms, and tables

ユースケース

01Cleaning and structuring legacy document data into JSON formats

02Automated data extraction from invoices, receipts, and forms

03High-volume parsing of standardized exam questions or quizzes

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add affaan-m/everything-claude-code regex-vs-llm-structured-text

For use in Claude.ai and ChatGPT

主な機能

0195%+ cost reduction compared to LLM-only pipelines

02Hybrid parsing architecture combining Regex and LLMs

03Programmatic confidence scoring to identify extraction errors

04Detailed metrics for success rates and pipeline health

05112,914 GitHub stars

06Implementation patterns for quizzes, forms, and tables

ユースケース

01Cleaning and structuring legacy document data into JSON formats

02Automated data extraction from invoices, receipts, and forms

03High-volume parsing of standardized exam questions or quizzes

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add affaan-m/everything-claude-code regex-vs-llm-structured-text

For use in Claude.ai and ChatGPT