> ## Documentation Index
> Fetch the complete documentation index at: https://wildcampstudio.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# What is Delve?

> AI-powered taxonomy generation using state-of-the-art language models

<Info>
  Delve helps you automatically generate taxonomies from unstructured data using Claude and other LLMs.
</Info>

## Overview

Delve is a production-ready SDK and CLI for automatically generating taxonomies from your data using state-of-the-art language models. It analyzes your documents, identifies patterns, and creates a structured taxonomy with automatic categorization.

Whether you have customer feedback, support tickets, user reviews, or any other unstructured text data, Delve can automatically discover categories and organize your content.

<Note>
  Delve is inspired by the [TNT-LLM paper](https://arxiv.org/abs/2403.12173), implementing a **hybrid unsupervised + supervised approach** that combines LLM-powered taxonomy discovery with efficient ML-based classification. [Learn more about the methodology →](/how-it-works)
</Note>

## Key Features

* **Automated Taxonomy Generation** - No manual category creation needed. Delve uses iterative minibatch-based clustering with Claude 3.5 Sonnet to automatically discover categories in your data.

* **Multiple Data Sources** - Work with CSV files, JSON/JSONL, LangSmith runs, or pandas DataFrames. Flexible adapters make it easy to process data from any source.

* **Smart Categorization** - Iterative refinement with minibatch clustering ensures high-quality taxonomies. Built-in LLM-based validation catches quality issues.

* **Flexible Exports** - Get your results in JSON, CSV, and Markdown reports. Machine-readable formats for integration, human-readable reports for analysis.

* **Both SDK and CLI** - Use Delve programmatically in your Python applications or from the command line for quick analysis.

* **Smart Sampling** - Automatically samples large datasets for efficient processing while maintaining representative coverage.

## How It Works

Delve uses a sophisticated multi-stage pipeline powered by LangGraph:

1. **Data Loading** - Adapters load data from various sources (CSV, JSON, LangSmith, DataFrame)
2. **Summarization** - Fast LLM generates concise summaries of each document to reduce token usage
3. **Minibatch Generation** - Documents are grouped into minibatches for efficient processing
4. **Iterative Clustering** - Each minibatch is analyzed to generate category candidates
5. **Taxonomy Update** - Categories are merged, refined, and consolidated across iterations
6. **Quality Review** - LLM validates taxonomy quality and completeness
7. **Document Labeling** - All documents are categorized with explanations
8. **Export** - Results saved in multiple formats (JSON, CSV, Markdown)

## Use Cases

Delve is perfect for:

* **Customer Feedback Analysis** - Automatically categorize feedback into product areas, features, and sentiment
* **Support Ticket Classification** - Organize support tickets by issue type, urgency, and category
* **Content Organization** - Create taxonomies for articles, documentation, or knowledge bases
* **Research Data Analysis** - Categorize survey responses, interview transcripts, or research notes
* **Social Media Monitoring** - Classify social media posts, comments, and mentions

## Next Steps

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Get started in 5 minutes
  </Card>

  <Card title="How It Works" icon="gears" href="/how-it-works">
    Understand the TNT-LLM methodology
  </Card>

  <Card title="Configuration Guide" icon="sliders" href="/configuration-guide">
    Tune parameters for your use case
  </Card>

  <Card title="Examples" icon="book" href="/examples">
    See code examples
  </Card>
</CardGroup>