About¶

The glyphdeck library is a comprehensive toolkit designed to streamline & simplify various aspects of asynchronous LLM data processing workflows over high-volume, dense semantic data - like customer feedback, reviews and comments.

Common aspects of the LLM data workflow are handled end-to-end including async llm calls, data validation, sanitisation, transformation, caching and step chaining, facilitating the fast development of robust, error free LLM data workflows.

Its also equiped with caching and a configurable logging facility that makes complex asyncronous LLM workflows much easier to configure, understand and debug.

Features¶

Asynchronous: Rip through tabular data with asynchronous llm usage
Auto-scaling: Don’t worry about api errors - the LLMHandler automatically scales to your api limits
Rapid config: Simple syntax in a highly abstracted interface makes set up a breeze
Data chaining: Automatically pass data between steps with the Cascade class
Self validating: The Cascade class automatically maintains dataset integrity
Pydantic models: Use built-in or custom Pydantic data models for structured llm output
Type enforcement: Get your structured data in the correct format from the llm every time
Private data: Sanitiser class strips out common private data patterns before making api calls
Automatic caching: Save on API costs and re-use results
Logging: Built-in logging facility helps to debug your set-ups & peer inside the event loop

Set-up¶

Install¶

pip install glyphdeck

Set API Key¶

Set your OPENAI_API_KEY with a command, or add it to your environment variables.

setx OPENAI_API_KEY "your_api_key_here"

Import¶

import glyphdeck as gd

Usage¶

Cascades¶

The Cascade class is the primary interface for the glyphdeck library. It handles and processes data in a record-like structure, providing easy to use syntax for LLM data handling workflows.

It validates and enforces all data movements against a common id, ensuring that each record has a unique, immutable identifier that remains consistent, regardless of other changes.

This all combines to let you spin up LLM workflows in record time.

Example¶

import glyphdeck as gd

# Provide a dataframe or a path to a file (csv or xlsx)
data_source = r"tests\testdata.pizzashopreviews.xlsx"

# Intialise cascade instance and identify the unique id (required) and target data
cascade = gd.Cascade(data_source, "Review", "Review Text")

# Optionally remove private information
cascade.sanitiser.run()

# Prepare the llm
cascade.set_llm_handler(
    provider="OpenAI",
    model="gpt-4o-mini",
    system_message=(
        "You are an expert pizza shop customer feedback analyst system."
        "Analyse the feedback and return results in the correct format."
    ),
    validation_model=gd.validators.SubCatsSentiment,
    cache_identifier="pizzshop_sentiment",
)

# Run the llm_handler
cascade.llm_handler.run("llm_category_sentiment")

Output¶

Review Id	Review	Review_sub_categories	Review_sentiment_score
1	I love this place, fantastic pizza and great service. It’s family run and it shows. Out of heaps of good local options it is still the…	Pizza Quality,Service Quality,Family Run,Local Options,Pricing,Parking	0.95
2	Got disappointed. Not good customer service. definitely not great pizza.We went there yesterday to…	Customer Service,Pizza Quality,Dine-In Experience,Promotions,Family Experience	-0.75

Tip

That’s just the start!

Glyphdeck is highly customisable ♾️ with useful optional commands in every function, class & method. Use the Cascade and other tools to mix and match, create feedback loops 🔁 and more - all with blazing 🔥 asynchronous speed through the OpenAI API.