Unstract, an LLM Data Extraction Platform for Unstructured Documents, Released

Unstract is an LLM-based platform that extracts necessary data from unstructured documents without a fixed format, such as contracts, invoices, and resumes, and converts it into structured JSON form.

Unlike existing regex- or template-based methods, users can define the fields and format to extract through natural language prompts, allowing flexible handling even when document formats change. The extracted data can be connected to existing applications via a REST API, or batch-processed into a data warehouse via an ETL pipeline.

Key Features and Architecture:

• Multi-LLM support: Supports OpenAI, Anthropic, Google Gemini, as well as local model integration via Ollama.
• Various connectors: Supports various storage options such as AWS S3 and Google Drive, vector DBs such as Qdrant and Pinecone, and various destinations such as Snowflake and BigQuery.
• Tech stack: Features a robust architecture based on React (frontend), Django (backend), Celery (asynchronous tasks), and PostgreSQL (database).
• Extensibility: Can be self-hosted via Docker Compose, and can also be connected to AI agents via an MCP (Model Context Protocol) server.