Chunkr provides a Document Intelligence API utilizing advanced VLMs and computer vision for high-accuracy data extraction from complex documents. The service transforms various file types, including those with tables, handwriting, and forms, into structured data like perfect HTML/Markdown with precise bounding boxes. It ensures schema compliance and provides multi-level citations with confidence metrics, supporting RAG-optimized workflows.
Funding
Funding not disclosed
Founders
Product
Problem
Extracting structured data from documents like invoices and receipts is a time-consuming and error-prone process when done manually. Existing solutions often lack the flexibility to handle diverse document types and complex layouts, leading to inaccurate data extraction and inefficient workflows.
Solution
Chunkr offers an open-source document intelligence API that converts complex documents into structured data ready for use in Retrieval-Augmented Generation (RAG) and Large Language Model (LLM) applications. The API leverages advanced layout analysis, OCR, and semantic chunking techniques to accurately identify and extract key information from PDFs, PPTs, Word documents, and images. Chunkr's architecture allows for segment-level processing, enabling users to configure parsing differently for titles, tables, formulas, and diagrams. Users can choose between fast OCR and VLM-powered extraction, and deploy via API or self-host in their VPC.
Target Audience
Chunkr is designed for developers and organizations building RAG applications, LLM-powered workflows, and document processing solutions across various industries, including finance, healthcare, and legal.
Features
- Layout analysis to identify titles, paragraphs, tables, and other document segments
- Multi-lingual OCR with word-level bounding boxes and auto text-layer detection
- VLM-powered parsing for tables, formulas, and custom parsing prompts
- Semantic chunking to maintain semantic integrity when splitting documents
- Flexible file handling for PDFs, PPTs, Word docs, images, URLs, and base64 encoded data
- Configurable pipelines with options for layout analysis and OCR providers
- Support for multiple LLM providers with customizable prompts and fallback mechanisms
- Built-in visibility dashboard to track ingest, view extraction results, and experiment with configurations
- Simple DevX with Python and curl integration
- Cloud-ready and self-hostable with Docker images and Helm charts