LLM Content Extractor

A robust content extractor for LLM outputs with support for extracting and parsing JSON, XML, HTML, and code blocks from raw strings.

✨ Features

• 🎯 Multiple Format Support: Extract JSON, XML, HTML, and code blocks
• 🛡️ Fault Tolerant:
  • Automatically handle Markdown code fences (```json ... ```)
  • Intelligently extract content embedded in text
  • Fix common LLM errors (e.g., trailing commas in JSON)
• 🏗️ Strategy Pattern: Easy to extend with custom extractors
• 📦 Simple API: Functional interface, ready to use
• 🧪 Well Tested: High test coverage for reliability
• 🔧 Type Safe: Full type annotations support

📦 Installation

Install with pip:

pip install llm-content-extractor

Install with Poetry:

poetry add llm-content-extractor

🚀 Quick Start

Basic Usage

from llm_content_extractor import extract, ContentType

# Extract JSON
json_text = '''
Here's the data you requested:
```json
{
    "name": "Alice",
    "age": 30,
    "hobbies": ["reading", "coding"],
}

'''

result = extract(json_text, ContentType.JSON) print(result) # {'name': 'Alice', 'age': 30, 'hobbies': ['reading', 'coding']}

### JSON Extraction Examples

```python
from llm_content_extractor import extract, ContentType

# 1. JSON with Markdown fence
text1 = '```json\n{"status": "success"}\n```'
extract(text1, ContentType.JSON)  # {'status': 'success'}

# 2. Plain JSON
text2 = '{"status": "success"}'
extract(text2, ContentType.JSON)  # {'status': 'success'}

# 3. JSON embedded in text
text3 = 'The result is: {"status": "success"} - done!'
extract(text3, ContentType.JSON)  # {'status': 'success'}

# 4. JSON with trailing commas (common LLM error)
text4 = '{"items": [1, 2, 3,],}'
extract(text4, ContentType.JSON)  # {'items': [1, 2, 3]}

# 5. Using string content type
extract(text1, "json")  # Also works

XML Extraction Examples

Given text containing a fenced XML block:

A response from the LLM:
```xml
<root>
    <item id="1">First</item>
    <item id="2">Second</item>
</root>

You can extract it with:
```python
from llm_content_extractor import extract, ContentType

# Assuming the text above is in a variable `xml_text`
result = extract(xml_text, ContentType.XML)
print(result)  # Returns cleaned XML string

HTML Extraction Examples

Given text containing a fenced HTML block:

LLM says:
```html
<div class="container">
    <h1>Title</h1>
    <p>Content here</p>
</div>

You can extract it with:
```python
from llm_content_extractor import extract, ContentType

# Assuming the text above is in a variable `html_text`
result = extract(html_text, ContentType.HTML)
print(result)  # Returns cleaned HTML string

Code Block Extraction Examples

1. Extract language-specific code

Given a Python code block:

```python
def greet(name):
    return f"Hello, {name}!"

print(greet("World"))

Extract it by specifying the language:
```python
from llm_content_extractor import extract, ContentType

# Assuming the text above is in a variable `python_code_text`
code = extract(python_code_text, ContentType.CODE, language='python')
print(code)
# Output:
# def greet(name):
#     return f"Hello, {name}!"
#
# print(greet("World"))

2. Extract any code block

Given a generic code block (no language specified):

</code></pre>
<p>const x = 42;
console.log(x);</p>
<pre><code>

Extract it without specifying a language:

# Assuming the text above is in a variable `generic_code_text`
code = extract(generic_code_text, ContentType.CODE)
print(code)  # const x = 42;\nconsole.log(x);

🎨 Advanced Usage

Using Extractor Classes Directly

from llm_content_extractor import JSONExtractor, XMLExtractor

# Use extractor classes directly
json_extractor = JSONExtractor()
result = json_extractor.extract('{"key": "value"}')

xml_extractor = XMLExtractor()
result = xml_extractor.extract('<root><item>test</item></root>')

Custom Extractors

Create custom extractors by inheriting from the ContentExtractor base class:

from llm_content_extractor.base import ContentExtractor
from llm_content_extractor import extract, ContentType, register_extractor
import json

class CustomJSONExtractor(ContentExtractor):
    def extract(self, raw_text: str):
        # Custom extraction logic
        cleaned = raw_text.strip()
        # ... your logic here
        return json.loads(cleaned)

# Register custom extractor
register_extractor(ContentType.JSON, CustomJSONExtractor)

# Use the custom extractor
result = extract(text, ContentType.JSON)

Using Custom Extractor Instances

from llm_content_extractor import extract, JSONExtractor

# Create a custom configured extractor
my_extractor = JSONExtractor(strict=True)

# Pass the extractor instance directly
result = extract(raw_text, ContentType.JSON, extractor=my_extractor)

🧪 Fault Tolerance Features

LLM Content Extractor handles various common issues in LLM outputs:

1. Markdown Code Fences

# ✅ Supports various fence formats
extract('```json\n{"a": 1}\n```', ContentType.JSON)
extract('```JSON\n{"a": 1}\n```', ContentType.JSON)  # Uppercase
extract('```\n{"a": 1}\n```', ContentType.JSON)      # No language identifier

2. Embedded Content

# ✅ Extract content from surrounding text
text = '''
Here is the configuration:
{"enabled": true, "timeout": 30}
This will set the timeout to 30 seconds.
'''
extract(text, ContentType.JSON)  # Successfully extracts

3. JSON Syntax Error Fixing

# ✅ Automatically fix trailing commas
extract('{"items": [1, 2,],}', ContentType.JSON)  # {'items': [1, 2]}
extract('[{"id": 1,}, {"id": 2,}]', ContentType.JSON)  # [{'id': 1}, {'id': 2}]

4. Nested Structures

# ✅ Handle complex nested structures
nested = {
    "user": {
        "profile": {
            "name": "Alice",
            "contacts": ["email", "phone"]
        }
    }
}
# Fully supported

🏗️ Architecture

This project uses the Strategy Pattern:

ContentExtractor (Abstract Base Class)
    ├── JSONExtractor
    ├── XMLExtractor
    ├── HTMLExtractor
    └── CodeBlockExtractor

This design provides:

• ✅ Easy to add new extractor types
• ✅ Single responsibility for each extractor
• ✅ Flexible replacement and extension of extraction logic

📚 API Reference

extract(raw_text, content_type, language="", extractor=None)

Main extraction function.

Parameters:

• raw_text (str): Raw string output from LLM
• content_type (ContentType | str): Content type (JSON, XML, HTML, CODE)
• language (str, optional): For CODE type, specify the programming language
• extractor (ContentExtractor, optional): Custom extractor instance

Returns:

• JSON: dict or list
• XML/HTML/CODE: str

Raises:

• ValueError: If valid content cannot be extracted
• TypeError: If an invalid extractor is provided

ContentType Enum

class ContentType(Enum):
    JSON = "json"
    XML = "xml"
    HTML = "html"
    CODE = "code"

Extractor Options

JSONExtractor

JSONExtractor(strict=False)

• strict: If True, disable auto-fixing of errors like trailing commas

XMLExtractor

XMLExtractor(validate=True, recover=True)

• validate: If True and lxml is available, validate XML syntax
• recover: If True, attempt to recover from malformed XML

HTMLExtractor

HTMLExtractor(validate=False, clean=False)

• validate: If True, validate HTML structure
• clean: If True, clean and normalize HTML

CodeBlockExtractor

CodeBlockExtractor(language="", strict=False)

• language: Specific language to extract (e.g., 'python', 'javascript')
• strict: If True, only extract fenced code blocks

🔧 Development

Setup

# Clone the repository
git clone https://github.com/aihes/llm-content-extractor.git
cd llm-content-extractor

# Install dependencies
poetry install

# Run tests
poetry run pytest

# Format code
poetry run black .

# Type checking
poetry run mypy llm_content_extractor

Running Tests

# Run all tests
poetry run pytest

# With coverage report
poetry run pytest --cov=llm_content_extractor --cov-report=html

# Run specific tests
poetry run pytest tests/test_json_extractor.py

📖 Publishing to PyPI

See docs/PUBLISHING.md for detailed publishing instructions.

Quick steps:

# 1. Update version
poetry version patch

# 2. Build
poetry build

# 3. Publish
poetry publish

🤝 Contributing

Contributions are welcome! Please follow these steps:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

💡 Use Cases

LLM Content Extractor is particularly useful for:

• 🤖 LLM Application Development: Extract structured data from model outputs
• 🔄 Data Pipelines: Clean and standardize AI-generated content
• 🧪 Testing Tools: Validate LLM output formats
• 📊 Data Processing: Batch process LLM responses

❓ FAQ

Q: Why is my JSON extraction failing?

A: Ensure the text contains valid JSON structure. This library tries multiple strategies, but cannot recover completely corrupted JSON.

Q: Can I extract multiple code blocks?

A: The current version extracts the first matching code block. To extract multiple blocks, use the extract_all_blocks() method on CodeBlockExtractor or call the function multiple times.

Q: Is there support for other formats?

A: Yes! You can add support for new formats by inheriting from ContentExtractor and registering it in the system.

Q: How do I enable strict mode?

A: Use the extractor classes directly:

extractor = JSONExtractor(strict=True)
result = extractor.extract(text)

🌟 Advanced Features

Language Detection

from llm_content_extractor.strategies import CodeBlockExtractor

extractor = CodeBlockExtractor()
code = "def hello(): return 'world'"
language = extractor.detect_language(code)  # Returns 'python'

Extract All Code Blocks

from llm_content_extractor.strategies import CodeBlockExtractor

extractor = CodeBlockExtractor()
blocks = extractor.extract_all_blocks(multi_code_text)
for block in blocks:
    print(f"{block['language']}: {block['code']}")

Validate XML/HTML

from llm_content_extractor.strategies import XMLExtractor, HTMLExtractor

xml_extractor = XMLExtractor()
is_valid = xml_extractor.is_valid_xml(xml_string)

html_extractor = HTMLExtractor()
is_valid = html_extractor.is_valid_html(html_string)

📚 Documentation

• Architecture - Detailed architecture documentation
• Publishing Guide - How to publish to PyPI
• Examples - Usage examples

🙏 Acknowledgments

Thanks to all contributors and developers using this project!

📬 Contact

• Report Issues: GitHub Issues
• Feature Requests: GitHub Discussions

---

If this project helps you, please consider giving it a ⭐️!