What types of data formats does LangChain support for processing?

LangChain supports multiple data formats to handle diverse use cases in building language model applications. The primary formats include plain text, structured data like JSON and CSV, documents such as PDFs and HTML, and database connections. These formats are processed through built-in components and integrations with external libraries, allowing developers to work with different data sources efficiently. LangChain’s flexibility ensures compatibility with common data types used in AI pipelines while abstracting low-level parsing tasks.

For structured data, LangChain provides tools to process JSON, CSV, and database records. JSON is widely used for APIs and configuration files, and LangChain can parse nested JSON structures to extract text or metadata for language models. CSV files are handled via integrations with libraries like pandas, enabling operations like filtering or aggregating tabular data before feeding it into a model. Database interactions are supported through ORM frameworks (e.g., SQLAlchemy) or direct SQL queries, letting developers retrieve rows as dictionaries or strings. For example, a developer might query a PostgreSQL table, convert results to plain text prompts, and generate summaries using LangChain’s chain components.

Unstructured data formats like PDFs, HTML, and plain text files are processed using document loaders. LangChain integrates with libraries such as PyPDF2 for PDF text extraction, Beautiful Soup for HTML parsing, and the Unstructured library for markdown or Word documents. These tools convert files into standardized Document objects containing text content and metadata. For instance, a PDF resume parsed via PyPDF2 can be split into sections (education, work experience) and used to answer queries about a candidate’s background. LangChain also supports web-based data via RSS feeds or API responses, which are often formatted as JSON or XML. By combining these integrations, developers can build pipelines that transform raw data into structured prompts or context for language models.