Markup

Markup is an annotation tool designed to transform unstructured free-text into structured datasets for natural language processing (NLP) and machine learning (ML) workflows. The platform leverages GPT-4 as an underlying engine to assist users in quickly defining and labeling entities, relationships, categories, or other schema elements within raw text. Rather than requiring manual tagging from scratch, Markup integrates an AI-assisted interface where users can set annotation rules, approve or correct automated suggestions, and export the resulting structured data in formats usable for training models or populating databases. The core function revolves around a “schema-first” approach. Users begin by designing a data model—for example, specifying fields such as person names, dates, locations, product attributes, or sentiment labels—that the tool should extract or classify from each piece of text. Once the schema is defined, texts (uploaded as plain files, CSV rows, or pasted content) are processed through GPT-4 powered extraction, which produces initial annotations. The user then enters a review and correction phase: a side-by-side interface shows the original text on the left and the extracted structured fields on the right, with highlights indicating where each value was found. Any misalignments or missing items can be adjusted with a few clicks, and the model learns from every correction to improve future extractions. Markup supports multiple annotation types: entity extraction (named entities like organization, time, currency), relation extraction (linking two entities with a predicate, e.g., “CEO of Company”), text classification (assigning a label to entire documents or paragraphs), and sequence labeling (e.g., part-of-speech tagging or custom spans). It also handl