Process Document with Chandra OCR (Datalab)

Runs a document through Chandra, Datalab's document OCR model, via their hosted conversion API. Chandra is oriented toward complex and academic documents: it handles intricate tables, equations, forms, and handwriting, preserves reading order, and extracts embedded figures/images as files.

Usage

chandra_ocr(
  file_path,
  api_key = Sys.getenv("DATALAB_API_KEY"),
  output_format = "markdown",
  mode = "balanced",
  use_llm = FALSE,
  langs = NULL,
  max_pages = NULL,
  output_file = NULL,
  timeout = 300
)

Arguments

file_path: Character string. Path to a local PDF, PNG, JPEG, or other supported document.
api_key: Character string. Datalab API key, sent in the X-Api-Key header. Default retrieves from environment variable "DATALAB_API_KEY".
output_format: Character string. One of "markdown" (default), "html", "json", or "chunks".
mode: Character string. Quality/speed tradeoff: "fast", "balanced" (default), or "accurate".
use_llm: Logical. If TRUE, adds an LLM pass to improve accuracy on tables, forms, and inline math (higher cost). Default is FALSE.
langs: Character string. Optional comma-separated OCR language hints (e.g. "English"). Default is NULL (auto).
max_pages: Integer. Optional cap on pages processed. Default is NULL.
output_file: Character string. Optional path to save the JSON response. Default is NULL (no file output).
timeout: Numeric. Maximum seconds to wait for the async job. Default is 300.

Value

List with structured_output containing the paginated markdown, output_format, page_count, images (named list of base64 strings), and metadata. Pass to chandra_extract_pages for a normalized page list.

Details

The Datalab API is asynchronous; this function submits the file and polls until the result is ready. See chandra_extract_pages to normalize the output into the per-page format shared across providers.

Author

Nathan C. Layman

Examples

if (FALSE) { # \dontrun{
# Key in DATALAB_API_KEY
result <- chandra_ocr("paper.pdf")
pages <- chandra_extract_pages(result)

# Higher accuracy with the LLM table pass
result <- chandra_ocr("paper.pdf", mode = "accurate", use_llm = TRUE)

# Access extracted figures (base64, keyed by filename)
figures <- result$structured_output$images
} # }