PDF Parsing Configuration

Arcana supports PDF document ingestion with pluggable parsers. The default uses Poppler’s pdftotext command-line tool.

Quick Start

# config/config.exs

# Default: Poppler's pdftotext
config :arcana, pdf_parser: :poppler

# With options
config :arcana, pdf_parser: {:poppler, layout: true}

# Custom parser
config :arcana, pdf_parser: MyApp.CustomPDFParser

Poppler Parser (Default)

The default PDF parser uses Poppler’s pdftotext command-line utility. It’s fast, reliable, and handles most PDFs well.

Installation

macOS
Ubuntu/Debian
Fedora
Docker

brew install poppler

sudo apt-get install poppler-utils

sudo dnf install poppler-utils

FROM elixir:1.16-alpine

# Install poppler-utils
RUN apk add --no-cache poppler-utils

# ... rest of your Dockerfile

Verify Installation

iex> Arcana.FileParser.PDF.Poppler.available?()
true

Configuration

# config/config.exs

# Use default settings
config :arcana, pdf_parser: :poppler

# Preserve original layout
config :arcana, pdf_parser: {:poppler, layout: true}

# Disable layout preservation for better text flow
config :arcana, pdf_parser: {:poppler, layout: false}

Options

Option	Default	Description
`:layout`	`true`	Preserve original text layout (columns, spacing)

Usage

# Ingest a PDF file
{:ok, document} = Arcana.ingest_file("path/to/document.pdf", repo: MyApp.Repo)

# With metadata
{:ok, document} = Arcana.ingest_file(
  "paper.pdf",
  repo: MyApp.Repo,
  metadata: %{"title" => "Research Paper", "author" => "Jane Doe"}
)

# Organize into collections
{:ok, document} = Arcana.ingest_file(
  "manual.pdf",
  repo: MyApp.Repo,
  collection: "manuals"
)

When to Use Layout Preservation

Layout: true
Layout: false

Best for:

Academic papers
Multi-column documents
Tables and structured data
Documents where spatial layout matters

config :arcana, pdf_parser: {:poppler, layout: true}

Output Example:

Column 1 text here       Column 2 text here
More col 1 content       More col 2 content

Best for:

Books and articles
Single-column documents
When you want continuous text flow
Better for chunking

config :arcana, pdf_parser: {:poppler, layout: false}

Output Example:

Column 1 text here More col 1 content
Column 2 text here More col 2 content

Custom PDF Parser

Implement the Arcana.FileParser.PDF behaviour for alternative PDF parsing solutions.

Implementation Examples

Apache PDFBox
Cloud API
Python PyPDF2
Rustler NIF

Using PDFBox via Java/Rustler:

defmodule MyApp.PDFBoxParser do
  @behaviour Arcana.FileParser.PDF

  @impl true
  def parse(path, opts) when is_binary(path) do
    # Call PDFBox CLI or Java interop
    case System.cmd("pdfbox", ["export:text", "-i", path]) do
      {text, 0} -> {:ok, text}
      {error, _} -> {:error, {:pdfbox_failed, error}}
    end
  end

  # Optional: declare binary support
  def supports_binary?, do: false
end

Configuration:

config :arcana, pdf_parser: MyApp.PDFBoxParser

Using a cloud PDF parsing service:

defmodule MyApp.CloudPDFParser do
  @behaviour Arcana.FileParser.PDF

  @impl true
  def parse(binary, opts) when is_binary(binary) do
    api_key = opts[:api_key] || System.get_env("PDF_API_KEY")
    
    # Upload to cloud service
    case HTTPoison.post(
      "https://api.pdfparse.com/v1/extract",
      binary,
      [
        {"Authorization", "Bearer #{api_key}"},
        {"Content-Type", "application/pdf"}
      ]
    ) do
      {:ok, %{body: body}} ->
        %{"text" => text} = Jason.decode!(body)
        {:ok, text}

      {:error, reason} ->
        {:error, reason}
    end
  end

  # Declare binary support for direct upload
  def supports_binary?, do: true
end

Configuration:

config :arcana, pdf_parser: {
  MyApp.CloudPDFParser,
  api_key: System.get_env("PDF_API_KEY")
}

Using PyPDF2 via Python port:

defmodule MyApp.PyPDFParser do
  @behaviour Arcana.FileParser.PDF

  @impl true
  def parse(path, _opts) when is_binary(path) do
    python_script = """
    import sys
    from PyPDF2 import PdfReader
    
    reader = PdfReader(sys.argv[1])
    text = ''
    for page in reader.pages:
        text += page.extract_text()
    print(text)
    """

    case System.cmd("python3", ["-c", python_script, path]) do
      {text, 0} -> {:ok, text}
      {error, _} -> {:error, {:pypdf_failed, error}}
    end
  end

  def supports_binary?, do: false
end

High-performance native PDF parsing:

defmodule MyApp.RustPDFParser do
  @behaviour Arcana.FileParser.PDF

  use Rustler,
    otp_app: :my_app,
    crate: :pdf_parser

  # NIF stub - implemented in Rust
  def parse_pdf_native(_path), do: :erlang.nif_error(:nif_not_loaded)

  @impl true
  def parse(path, _opts) when is_binary(path) do
    case parse_pdf_native(path) do
      {:ok, text} -> {:ok, text}
      {:error, reason} -> {:error, reason}
    end
  end

  def supports_binary?, do: false
end

Binary Content Support

Some parsers can process PDF binary content directly (useful for file uploads):

defmodule MyApp.BinaryPDFParser do
  @behaviour Arcana.FileParser.PDF

  @impl true
  def parse(binary, opts) when is_binary(binary) do
    # Handle both file paths and binary content
    if File.exists?(binary) do
      # It's a file path
      File.read!(binary) |> extract_text(opts)
    else
      # It's binary content
      extract_text(binary, opts)
    end
  end

  # Declare binary support
  def supports_binary?, do: true

  defp extract_text(binary, _opts) do
    # Your PDF extraction logic
    {:ok, "extracted text"}
  end
end

Check if a parser supports binary input:

iex> Arcana.FileParser.PDF.supports_binary?({MyApp.BinaryPDFParser, []})
true

iex> Arcana.FileParser.PDF.supports_binary?({Arcana.FileParser.PDF.Poppler, []})
false

Parser Configuration

Global Configuration

# config/config.exs

# Use default Poppler
config :arcana, pdf_parser: :poppler

# Custom parser with options
config :arcana, pdf_parser: {
  MyApp.CloudPDFParser,
  api_key: System.get_env("PDF_API_KEY"),
  timeout: 30_000
}

Per-Call Override

# Override parser for specific files
Arcana.ingest_file(
  "complex.pdf",
  repo: MyApp.Repo,
  pdf_parser: MyApp.AdvancedParser
)

# With custom options
Arcana.ingest_file(
  "scanned.pdf",
  repo: MyApp.Repo,
  pdf_parser: {MyApp.OCRParser, language: "eng"}
)

Advanced Parsing

OCR for Scanned PDFs

Handle scanned PDFs with OCR:

defmodule MyApp.OCRParser do
  @behaviour Arcana.FileParser.PDF

  @impl true
  def parse(path, opts) do
    language = opts[:language] || "eng"
    
    # First try normal text extraction
    case Arcana.FileParser.PDF.Poppler.parse(path, opts) do
      {:ok, text} when byte_size(text) > 100 ->
        # Got enough text, no OCR needed
        {:ok, text}

      _ ->
        # Fallback to OCR
        perform_ocr(path, language)
    end
  end

  defp perform_ocr(path, language) do
    # Use Tesseract via ImageMagick + Tesseract
    case System.cmd("sh", ["-c", "
      pdftoppm #{path} page -png | 
      tesseract stdin stdout -l #{language}
    "]) do
      {text, 0} -> {:ok, text}
      {error, _} -> {:error, {:ocr_failed, error}}
    end
  end

  def supports_binary?, do: false
end

Extracting Tables

Extract structured data from PDF tables:

defmodule MyApp.TableParser do
  @behaviour Arcana.FileParser.PDF

  @impl true
  def parse(path, opts) do
    # Use tabula-py or camelot-py for table extraction
    case System.cmd("python3", ["-c", "
      import tabula
      tables = tabula.read_pdf('#{path}', pages='all')
      text = '\\n'.join([df.to_string() for df in tables])
      print(text)
    "]) do
      {text, 0} -> {:ok, text}
      {error, _} -> {:error, {:table_extraction_failed, error}}
    end
  end

  def supports_binary?, do: false
end

Metadata Extraction

Extract PDF metadata along with text:

defmodule MyApp.MetadataPDFParser do
  @behaviour Arcana.FileParser.PDF

  @impl true
  def parse(path, opts) do
    with {:ok, text} <- extract_text(path),
         {:ok, metadata} <- extract_metadata(path) do
      # Include metadata in the text for ingestion
      enriched_text = """
      Title: #{metadata[:title]}
      Author: #{metadata[:author]}
      Created: #{metadata[:created]}

      #{text}
      """
      
      {:ok, enriched_text}
    end
  end

  defp extract_text(path) do
    Arcana.FileParser.PDF.Poppler.parse(path, [])
  end

  defp extract_metadata(path) do
    case System.cmd("pdfinfo", [path]) do
      {info, 0} ->
        metadata = parse_pdfinfo(info)
        {:ok, metadata}

      _ ->
        {:ok, %{}}
    end
  end

  defp parse_pdfinfo(info) do
    info
    |> String.split("\n")
    |> Enum.reduce(%{}, fn line, acc ->
      case String.split(line, ":", parts: 2) do
        [key, value] ->
          Map.put(acc, String.downcase(key) |> String.to_atom(), String.trim(value))

        _ ->
          acc
      end
    end)
  end

  def supports_binary?, do: false
end

Testing PDF Parsers

defmodule MyApp.PDFParserTest do
  use ExUnit.Case

  @fixtures_path "test/fixtures/pdfs"

  test "parses simple PDF" do
    path = Path.join(@fixtures_path, "simple.pdf")
    {:ok, text} = MyApp.CustomPDFParser.parse(path, [])

    assert text =~ "expected content"
    assert String.length(text) > 100
  end

  test "handles multi-page PDFs" do
    path = Path.join(@fixtures_path, "multipage.pdf")
    {:ok, text} = MyApp.CustomPDFParser.parse(path, [])

    assert text =~ "page 1 content"
    assert text =~ "page 2 content"
  end

  test "returns error for corrupted PDF" do
    path = Path.join(@fixtures_path, "corrupted.pdf")
    assert {:error, _} = MyApp.CustomPDFParser.parse(path, [])
  end

  test "supports binary content" do
    path = Path.join(@fixtures_path, "simple.pdf")
    binary = File.read!(path)

    if MyApp.CustomPDFParser.supports_binary?() do
      {:ok, text} = MyApp.CustomPDFParser.parse(binary, [])
      assert text =~ "expected content"
    end
  end
end

Error Handling

defmodule MyApp.RobustPDFParser do
  @behaviour Arcana.FileParser.PDF

  @impl true
  def parse(path, opts) do
    cond do
      not File.exists?(path) ->
        {:error, :file_not_found}

      not pdf_file?(path) ->
        {:error, :not_a_pdf}

      true ->
        do_parse(path, opts)
    end
  end

  defp pdf_file?(path) do
    # Check magic number
    case File.read(path, 4) do
      {:ok, <<"%PDF">>} -> true
      _ -> false
    end
  end

  defp do_parse(path, opts) do
    timeout = opts[:timeout] || 30_000

    Task.async(fn ->
      Arcana.FileParser.PDF.Poppler.parse(path, opts)
    end)
    |> Task.await(timeout)
  rescue
    exception ->
      {:error, {:parse_failed, exception}}
  catch
    :exit, reason ->
      {:error, {:timeout, reason}}
  end

  def supports_binary?, do: false
end

Best Practices

Start with Poppler - Works well for most PDFs
Add OCR for scanned docs - Fallback when text extraction fails
Set timeouts - Large PDFs can take time to parse
Validate input - Check file exists and is actually a PDF
Handle errors gracefully - Log failures but don’t crash
Test with real PDFs - Different PDF generators produce different output
Consider layout preservation - Enable for multi-column, disable for books

Troubleshooting

pdftotext not found

Install Poppler:

# macOS
brew install poppler

# Ubuntu
sudo apt-get install poppler-utils

Verify installation:

which pdftotext

Extracted text is garbled

Try disabling layout preservation:

config :arcana, pdf_parser: {:poppler, layout: false}

Or use a different parser for problematic PDFs.

Scanned PDFs return empty text

Use OCR:

config :arcana, pdf_parser: MyApp.OCRParser

Requires Tesseract:

brew install tesseract  # macOS
apt-get install tesseract-ocr  # Ubuntu

PDF parsing is slow

Implement async parsing with timeouts:

Task.async(fn -> parse_pdf(path) end)
|> Task.await(30_000)

Or use a faster parser like Rust-based solutions.

Next Steps

Chunkers

Configure text splitting after parsing

LLM Integration

Setup LLMs for question answering

Getting Started

Core Concepts

Guides

Configuration

PDF Parsing Configuration

Quick Start

Poppler Parser (Default)

Installation

Verify Installation

Configuration

Options

Usage

When to Use Layout Preservation

Custom PDF Parser

Implementation Examples

Binary Content Support

Parser Configuration

Global Configuration

Per-Call Override

Advanced Parsing

OCR for Scanned PDFs

Extracting Tables

Metadata Extraction

Testing PDF Parsers

Error Handling

Best Practices

Troubleshooting

Next Steps

Chunkers

LLM Integration

Build docs developers (and LLMs) love

Getting Started

Core Concepts

Guides

Configuration

​Quick Start

​Poppler Parser (Default)

​Installation

​Verify Installation

​Configuration

​Options

​Usage

​When to Use Layout Preservation

​Custom PDF Parser

​Implementation Examples

​Binary Content Support

​Parser Configuration

​Global Configuration

​Per-Call Override

​Advanced Parsing

​OCR for Scanned PDFs

​Extracting Tables

​Metadata Extraction

​Testing PDF Parsers

​Error Handling

​Best Practices

​Troubleshooting

​Next Steps

Chunkers

LLM Integration

Build docs developers (and LLMs) love

Quick Start

Poppler Parser (Default)

Installation

Verify Installation

Configuration

Options

Usage

When to Use Layout Preservation

Custom PDF Parser

Implementation Examples

Binary Content Support

Parser Configuration

Global Configuration

Per-Call Override

Advanced Parsing

OCR for Scanned PDFs

Extracting Tables

Metadata Extraction

Testing PDF Parsers

Error Handling

Best Practices

Troubleshooting

Next Steps