This Langflow flow creates an intelligent document routing and ingestion system that automatically categorizes uploaded files and stores them in separate knowledge bases.

The flow starts by reading files through a File component, which can handle various document formats including PDFs, Word documents, images, and text files. These files are then passed to a Smart Router component that uses a language model to automatically categorize the content. The router is configured with two routes: "HR" for resumes and HR-related documents, and "Legal" for contracts and agreements. The language model analyzes each document's content and determines which category it belongs to based on the routing instructions.

Once categorized, the documents follow separate processing paths. Each route feeds into its own Split Text component that breaks the documents into manageable chunks with configurable size and overlap settings. The chunked text is then processed by dedicated Knowledge Ingestion components that store the data in separate knowledge bases - "HR_BASE" for HR documents and "Legal_BASE" for legal documents. Each knowledge base is configured to vectorize the text content for semantic search capabilities while maintaining the original document structure and metadata.

The classification process relies on either a Prompt Template combined with a Language Model or the Structured Output component to extract document types with confidence scores. The routing logic uses conditional components to evaluate classifications and direct documents to appropriate processing branches. Files can be uploaded through the UI, programmatically via the v2 files API, or attached to chat and webhook triggers, with support for files up to 1024 MB in common formats.

The flow can be extended using additional Langflow components for more sophisticated processing. Teams can add confidence-aware routing by incorporating confidence scores from Structured Output and sending uncertain classifications to human reviewers via Slack integration. Storage options can be customized with S3 uploaders or local file storage, while database logging captures classification results. Multi-step processing chains can trigger secondary flows for OCR, redaction, or document signing workflows. For more complex scenarios, Agent components can replace simple language models to enable dynamic decision-making with access to external tools and APIs.