Deep-TOON: Deep Token-Oriented Object Notation

Deep-TOON is a token-optimized JSON representation format designed for LLMs and AI applications. It provides significant compression for nested JSON structures while maintaining perfect data fidelity and LLM readability.

📊 Performance Overview

Test Data: dummyjson.com/users (3 users)

Original JSON:    1,675 tokens
Deep-TOON:       1,065 tokens (36.4% reduction)

Comprehensive Test Results:

• Average reduction: 28.7% across diverse data types
• Best case: 61.0% reduction on large structured datasets
• Success rate: 92.9% perfect roundtrip fidelity

🏗️ Format Specification

Basic Structure

[N,delimiter]{schema}:
  value1,value2,value3
  value4,value5,value6

Hierarchical Tuples

Deep-TOON uses explicit hierarchical notation to group related fields:

# Nested objects become tuples
address{street,city,coordinates{lat,lng}}

# Results in data like:
("626 Main Street", "Phoenix", (-77.16, -92.08))

Complete Example

Original JSON:

{
  "users": [
    {
      "id": 1,
      "firstName": "Emily", 
      "lastName": "Johnson",
      "age": 28,
      "address": {
        "address": "626 Main Street",
        "city": "Phoenix", 
        "state": "Mississippi",
        "coordinates": {"lat": -77.16213, "lng": -92.084824}
      },
      "bank": {
        "cardNumber": "9289760655481815",
        "cardType": "Elo"
      }
    }
  ],
  "total": 208,
  "skip": 0,
  "limit": 3
}

Deep-TOON Format:

users[1,]{id,firstName,lastName,age,address{address,city,state,coordinates{lat,lng}},bank{cardNumber,cardType}}:
  1,Emily,Johnson,28,("626 Main Street",Phoenix,Mississippi,(-77.16213,-92.084824)),("9289760655481815",Elo)
total: 208
skip: 0  
limit: 3

🔧 Usage Examples

Installation

pip install deep-toon

Basic Usage

import deep_toon

# Your JSON data
data = {
    "users": [
        {
            "id": 1,
            "name": "Alice",
            "address": {
                "street": "123 Main St",
                "city": "NYC",
                "coordinates": {"lat": 40.7, "lng": -74.0}
            }
        },
        {
            "id": 2,
            "name": "Bob", 
            "address": {
                "street": "456 Oak Ave",
                "city": "LA", 
                "coordinates": {"lat": 34.0, "lng": -118.2}
            }
        }
    ]
}

# Compress to Deep-TOON format
compressed = deep_toon.encode(data)
print("Compressed:", compressed)

# Decompress back to original
original = deep_toon.decode(compressed)
print("Original data restored:", data == original)

Output:

users[2,]{id,name,address{street,city,coordinates{lat,lng}}}:
  1,Alice,("123 Main St",NYC,(40.7,-74.0))
  2,Bob,("456 Oak Ave",LA,(34.0,-118.2))

Advanced Usage

# Use the classes directly for more control
from deep_toon import DeepToonEncoder, DeepToonDecoder

encoder = DeepToonEncoder()
decoder = DeepToonDecoder()

# Custom delimiter for data with commas
encoder = DeepToonEncoder(delimiter=';')
compressed = encoder.encode(data)

Smart Encoding (Save-Safe)

Use smart_encode to automatically fall back to minified JSON if Deep-TOON doesn't achieve a specified savings threshold (default 10%).

from deep_toon import smart_encode

# Only use Deep-TOON if it saves > 10% tokens
# Otherwise returns minified JSON
encoded = smart_encode(data, threshold=0.1)

# You can also use a custom token counter (defaults to char length)
encoded = smart_encode(data, token_counter=len)

🎨 Format Features

Schema Declaration

The schema explicitly declares the structure:

{field1,field2,nested{subfield1,subfield2},deep{level1{level2}}}

Tuple Nesting

Related fields are grouped into tuples:

# Person with address
person{name,age,address{street,city}}
# Results in: ("Alice", 30, ("123 Main", "NYC"))

Null Handling

Missing or null values are handled gracefully:

# With missing city
("123 Main", null, (40.7, -74.0))

Quoting Rules

Strings are quoted only when necessary:

# No quotes needed
Simple,Text,123

# Quotes for special characters  
"Text with, comma","Multi word text","123-abc"

🎨 Deep-TOON Design Philosophy

Deep-TOON uses hierarchical tuples to represent nested structures efficiently:

// Original JSON
{"user": {"profile": {"name": "Alice", "age": 30}}}

// Deep-TOON representation
[1,]{user{profile{name,age}}}:
  (("Alice",30))

Key Benefits:

1. Compact schemas - Structure declared once, no repetition
2. Explicit hierarchy - Clear nesting with {...} notation
3. Tuple efficiency - Related data grouped logically
4. LLM optimized - Easy to read and parse

🚀 Performance Characteristics

When Deep-TOON Excels

• Nested objects (addresses, preferences, metadata)
• Repeated structures (arrays of complex objects)
• Deep hierarchies (API responses, config files)
• Mixed data types (numbers, strings, booleans together)

Token Savings by Data Type

Data Type	Typical Reduction
Flat objects	10-30%
1-level nesting	25-45%
2+ level nesting	30-60%
Array of objects	35-50%

🔧 Advanced Usage

Custom Delimiters

# Use semicolon delimiter for data containing commas
encoder = DeepToonEncoder(delimiter=";")

Handling Large Arrays

# Deep-TOON automatically detects when arrays are worth compressing
# Arrays with <2 items or inconsistent schemas fall back to JSON

Error Handling

try:
    decoded = decoder.decode(deep_toon_string)
except DeepToonDecodeError as e:
    print(f"Decode error: {e}")
    # Handle malformed Deep-TOON data

📈 Use Cases

• LLM Training Data - Reduce token costs for large datasets
• API Response Compression - Faster transmission and processing
• Configuration Files - More readable than JSON for complex configs
• Data Interchange - Efficient format for AI-to-AI communication
• Prompt Engineering - Include more context in limited token budgets

🔬 Technical Details

Schema Detection Algorithm

1. Field Analysis - Identify primitive vs nested fields
2. Structure Grouping - Group related fields into tuples
3. Optimization - Choose best compression strategy per field group
4. Schema Generation - Create hierarchical schema notation

Parsing Strategy

1. Pattern Matching - Detect Deep-TOON tabular format
2. Schema Parsing - Build nested structure from schema
3. Smart Splitting - Handle quoted strings and nested tuples
4. Type Inference - Convert strings back to appropriate types

🤝 Contributing

Deep-TOON is designed to be extended and improved. Key areas for contribution:

• Performance optimization for very large datasets
• Additional encoding strategies for specific data patterns
• Language bindings for other programming languages
• Integration tools for popular APIs and frameworks

📄 License

Apache 2.0 License - Free for commercial and personal use!

---

Deep-TOON - Efficient JSON representation for LLM applications. 🚀✨