New Plugins | Connecting Enterprise Documents to AI with DolphinDB

AI applications are moving beyond simple Q&A toward knowledge retrieval, data analysis, and complex task execution. At the same time, a growing share of enterprise knowledge lives in unstructured formats such as documents and text. How can this information be brought into data processing workflows and made accessible to AI models?

DolphinDB has introduced two new plugins, DocParser and ModelInference, to address this need. DocParser handles document parsing, while ModelInference provides model inference and embedding capabilities. Together, they connect the pipeline from documents to data, and from data to models.

DocParser: Making Enterprise Documents AI-Ready

Enterprise knowledge isn't stored only in databases. Research reports, product manuals, business policies, and project proposals are often kept in PDF, Word, PowerPoint, Excel, and TXT files. Before AI can use this information, the first challenge is not how to query it, but how to extract its content and turn it into structured data.

DocParser supports common document formats, including PDF, DOCX, PPTX, XLSX, and TXT, and converts raw documents into structured text ready for downstream processing. In RAG (Retrieval-Augmented Generation) scenarios, this means knowledge scattered across different files can first be turned into data that retrieval workflows can work with.

ModelInference: Bringing Model Inference into Data Workflows

Document parsing is only the first step in a RAG workflow. The extracted text must also be converted into vector representations that support semantic comparison and retrieval.

The ModelInference plugin brings model inference directly into DolphinDB's data processing workflows. Its current release focuses on embedding: text can be converted into FLOAT vectors for vector storage, similarity search, retrieval, and ranking. These capabilities can support semantic search, knowledge bases, and RAG applications. ModelInference plans to add further inference tasks in future releases, including time-series forecasting.

Two Plugins, One RAG Data Pipeline

DocParser and ModelInference cover two connected stages of the AI data pipeline:

• DocParser turns documents into processable data.
• ModelInference turns text into vector representations for semantic search.

In a typical RAG workflow, DocParser first parses documents such as PDF, Word, PowerPoint, and Excel files. DolphinDB then cleans the extracted content and splits it into chunks, and ModelInference generates an embedding for each chunk.

The resulting vectors are stored in DolphinDB's VectorDB, where vector indexes support similarity search, retrieval, and ranking. Meanwhile, the original text and related metadata can be stored in TextDB for full-text search and other data processing tasks.

From document parsing and text processing to embedding, storage, and retrieval, DolphinDB provides an integrated data foundation for RAG and other AI applications.

Expanding DolphinDB's AI Capabilities

DocParser and ModelInference are the latest steps in the evolution of DolphinDB's AI capabilities. DolphinDB has steadily expanded its AI stack, starting with built-in distributed machine learning operators, then adding SVM and XGBoost plugins, and later integrating with external deep learning frameworks through LibTorch and AI Dataloader. The stack now also includes DolphinMind for intelligent Q&A and DolphinX for AI agent development and governance.

With these two plugins, DolphinDB extends further into AI data processing and model inference, allowing it to work not only with structured data, but also with unstructured documents, enterprise knowledge, and models. For enterprises building AI-powered applications, this means a more unified foundation from raw documents to model-ready data.