Unstructured Platform

We offer multiple Source Connectors. We can connect to your data in its existing location.

Routing determines which strategy we will employ in transforming your document to our canonical JSON schema. There are three Partioning Strategies for document transformation, fast, hires, or ocr_only. fast is great for when there is extractable text available, like in HTML files or in the Microsoft Office Document format. hires is best for PDFs and tables and where accurate classification of document elements is critical. ocr_only is useful when dealing with image-based files or PDFs that do not have extractable text. If you’re unsure, select auto and we’ll handle the decision for you.

Your source document is transformed to our canonical JSON schema. Irrespective of the input document, the JSON schema we provide gives you a standardized output to code against. It contains 20+ elements, such as Header, Footer, Title, NarrativeText, Table, Image, and more. Each document is wrapped in extensive metadata so you can understand languages, file_type, source, hierarchy and much more.

Initially platform comes with two chunking strategies. Basic: Combines sequential elements up to specified size limits. Oversized elements are split, while tables are isolated and divided if necessary. Overlap between chunks is optional. By Title: Semantic chunking, understands the layout of the document and makes intelligent splits.

Call out to third party embedding providers, Open AI, AWS Bedrock, and Octo ML.

We have multiple Destination Connectors. Including all major vector databases.