Use Cases

Optimizing Data for RAG Systems

Raw text extraction often yields poor-quality chunks for Retrieval-Augmented Generation (RAG) by splitting sentences awkwardly or merging unrelated sections. Layout-aware parsing identifies logical content blocks (paragraphs, list items, table cells, headers). This enables the creation of cleaner, semantically coherent chunks, significantly improving the quality of context retrieved for LLMs.

• Example: Parsing a multi-column PDF ensures that text within a single column block is treated as a coherent unit, preventing it from being jumbled with text from adjacent columns during chunking, leading to more accurate RAG results.

Structured Data Extraction from Documents

Extract specific information, not just plain text. Use the parser's ability to identify element types (titles, tables, lists, key-value pairs inferred from layout) to pull structured data from documents like invoices, reports, forms, or contracts for automation or analysis.

• Example: Automatically extracting all tables from a collection of PDF financial statements, using the parser's structured output to identify table boundaries and cells, then converting this data into CSV or JSON for ingestion into an analytics platform.

Powering Document Understanding Pipelines

Use the Parsing API as the critical first step in sophisticated document processing workflows. The structured, layout-aware output provides vital context (element type, location, relationships) needed for subsequent AI tasks like document classification, named entity recognition (NER), summarization, or clause identification in legal documents.

• Example: A contract review system first uses the Parsing API to segment a DOCX file into distinct clauses (identified as structured paragraphs or sections). Then, a separate classification model analyzes the text of each identified clause to categorize it (e.g., 'Limitation of Liability', 'Payment Terms').

Content Migration and Normalization

When migrating content from diverse and complex formats (PDFs, DOCX, legacy HTML) into a unified system (like a modern CMS, knowledge base, or vector store), use the Parsing API to convert them into a consistent, clean, structured format (e.g., Markdown or structured JSON) while preserving important semantic structure like headings, lists, and tables.

• Example: Converting thousands of historical project reports stored as PDFs and Word documents into clean, searchable Markdown for a new internal knowledge portal, ensuring that document titles become H1 tags, sections become H2/H3 tags, and bullet points are correctly formatted as lists.

Last updated: May 2, 2025