Overview

PXL (Probably Expression Language) is a focused language with six core functions for data analysis: filter, extract, classify, score, ntile, and word_count.

PXL provides six essential functions for data analysis:

• filter: Apply conditional filters to data
• extract: Use AI to extract themes and concepts from text
• classify: Categorize text into defined groups
• score: Rate text along semantic scales
• ntile: Create statistical buckets from numeric data
• word_count: Count words in text fields

PXL is designed to be:

• Simple: Six functions cover most data analysis needs
• Readable: Functions use natural language descriptions
• AI-Powered: Text functions leverage machine learning
• Focused: Built specifically for analytical workflows

filter(revenue > 1000 and status = "active")

classify(customer_feedback into ("positive", "negative", "neutral"))

filter(amount > 500)
extract("sentiment" from review_text)
score(review_text from "negative" to "positive")

Text Analysis Functions

• extract: Uses AI clustering to identify themes in text
• classify: AI-powered categorization into defined groups
• score: Semantic rating along specified scales
• Natural language descriptions in function parameters

Available Functions

// Filter data based on conditions
filter(column_name > value and other_column = "text")

// Extract information using AI
extract("what to extract" from text_column)

// Classify data into categories
classify(text_column into ("category1", "category2", "category3"))

// Score text between two concepts
score(text_column from "negative" to "positive")

// Statistical bucketing
ntile(numeric_column, 4)

// Count words in text
word_count(text_column)

Individual Functions Each PXL function operates on data independently:

filter(status = "active")
classify(feedback into ("positive", "negative"))
score(feedback from "bad" to "good")

Semantic Operations

• Extract: Pull specific information from text using AI
• Classify: Categorize text into predefined groups
• Score: Rate text on semantic scales
• Filter: Apply AI-powered filtering criteria

Example

extract("customer sentiment" from review_text) ->
classify(review_text into ("complaint", "praise", "question")) ->
score(review_text from "negative" to "positive")

Statistical Bucketing

ntile(revenue, 5)  // Split into 5 revenue tiers
word_count(description)  // Count words in text

Filtering Examples

filter(amount > 1000 and status = "active")
filter(category in ("A", "B", "C"))
filter(date after "2023-01-01")

DuckDB Integration

• Functions execute using DuckDB’s query engine
• Columnar processing for analytical workloads
• Automatic parallelization for larger datasets
• Memory-efficient processing

• Text functions use clustering to reduce API calls
• Caching of similar text analysis results
• Efficient batching for large text datasets

Expression Bar

• Type PXL expressions in the formula bar
• Apply functions to create new columns
• View function results in spreadsheet
• Error messages for invalid syntax

Question-Based Analysis

• Ask questions that use PXL functions behind the scenes
• Agent automatically applies appropriate functions
• Text analysis through natural language queries
• Filter and analyze data through conversation

1. Identify Analysis Need: Determine what function to use
2. Write Expression: Use appropriate PXL function
3. Apply to Data: Execute function on dataset
4. Review Results: Check output and refine if needed

Function Usage

// Use descriptive text in AI functions
extract("customer sentiment" from review_text)
classify(feedback into ("complaint", "praise", "question"))
score(review_text from "very negative" to "very positive")

// Filter data efficiently
filter(status = "active" and amount > 1000)

Effective Analysis

• Use specific descriptions in text functions
• Filter data to focus analysis
• Check results and refine function parameters
• Combine different function types as needed