Elsai Text Extractors

Package: elsai-text-extractors  v0.1.0

Extracts text content from PDF, DOCX, CSV, Excel, and other document formats. Each extractor takes the file path at construction time and exposes a dedicated extraction method.

For LLM-powered natural language querying over Excel files, see ExcelParser (separate package).

Installation

pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-text-extractors/ elsai-text-extractors==0.1.0

Requirements: Python >= 3.9

---

Available extractors

Class	Package	Input format	Method
CSVFileExtractor	elsai-text-extractors	CSV	load_from_csv()
DocxTextExtractor	elsai-text-extractors	DOCX	extract_text_from_docx()
PyPDFTextExtractor	elsai-text-extractors	PDF	extract_text_from_pdf()
UnstructuredExcelLoaderService	elsai-text-extractors	XLSX	load_excel()
MarkItDownTextExtractor	elsai-text-extractors	PDF, DOCX, PPTX, and more	extract_text_from_file()
ExcelParser	elsai-parsers	XLSX (LLM-powered)	parse()

---

CSVFileExtractor

Loads and reads content from CSV files. The file path is passed at construction time.

from elsai_text_extractors.csv_extractor import CSVFileExtractor

extractor = CSVFileExtractor(file_path="sample_data/sample.csv")
content = extractor.load_from_csv()
print(content)

Constructor parameters:

Parameter	Description
file_path	Path to the CSV file

---

DocxTextExtractor

Extracts plain text from Microsoft Word (.docx) files.

from elsai_text_extractors.docx_extractor import DocxTextExtractor

extractor = DocxTextExtractor(file_path="sample_data/sample.docx")
content = extractor.extract_text_from_docx()
print(content)

Constructor parameters:

Parameter	Description
file_path	Path to the .docx file

---

PyPDFTextExtractor

Extracts text from PDF files using PyPDF.

from elsai_text_extractors.pypdfloader import PyPDFTextExtractor

extractor = PyPDFTextExtractor(file_path="sample_data/sample.pdf")
content = extractor.extract_text_from_pdf()
print(content)

Constructor parameters:

Parameter	Description
file_path	Path to the PDF file

---

UnstructuredExcelLoaderService

Loads content from Excel (.xlsx) files using the Unstructured library.

from elsai_text_extractors.unstructured_excel_loader import UnstructuredExcelLoaderService

extractor = UnstructuredExcelLoaderService(file_path="sample_data/sample.xlsx")
content = extractor.load_excel()
print(content)

Constructor parameters:

Parameter	Description
file_path	Path to the .xlsx file

---

MarkItDownTextExtractor

Converts a wide range of document formats into clean Markdown text using the MarkItDown library. Useful when you need consistent text output across mixed document types.

Supported formats: PDF, DOCX, PPTX, XLSX, and other formats supported by MarkItDown.

from elsai_text_extractors.markitdown import MarkItDownTextExtractor

extractor = MarkItDownTextExtractor(file_path="/path/to/document.pdf")
content = extractor.extract_text_from_file()
print(content)

Constructor parameters:

Parameter	Description
file_path	Path to the document file

---

ExcelParser

Package: elsai-parsers  v0.1.0

Queries Excel files using natural language. Unlike the extractors above which return raw text, ExcelParser indexes the Excel data into a vector database and uses an LLM to answer questions about it — including returning structured JSON responses.

pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-parsers/ elsai-parsers==0.1.0

from elsai_parsers.excel_parser import ExcelParser
from elsai_model.azure_openai import AzureOpenAIConnector
from elsai_embeddings.azure_embeddings import AzureOpenAIEmbeddingModel
from elsai_vectordb.chromadb import ChromaVectorDb

llm = AzureOpenAIConnector(...)
embedding_model = AzureOpenAIEmbeddingModel(...)
vector_db = ChromaVectorDb(persist_directory="./excel_db")

parser = ExcelParser(
    file_path="data/sales_report.xlsx",
    config={
        "vector_database": vector_db,
        "embedding_function": embedding_model,
        "llm": llm,
        "vector_store_type": "chroma",
    },
)

# Natural language query — returns plain text
result = parser.parse(user_prompt="What is the total revenue for Q3?")
print(result)

# With a JSON template — returns structured dict
result = parser.parse(
    user_prompt="List the top 5 products by sales.",
    json_template='{"products": [{"name": "", "sales": 0}]}',
)
print(result)

Constructor parameters:

Parameter	Description
file_path	Path to the Excel (.xlsx) file
config	Configuration dictionary (see below)

config dictionary keys:

Key	Description
vector_database	Vector database instance for indexing Excel content (e.g. ChromaVectorDb)
embedding_function	Embedding model instance for vectorizing content
llm	LLM connector instance for answering queries
vector_store_type	String identifier for the vector store type (e.g. "chroma")

parse() parameters:

Parameter	Description
user_prompt	Natural language question about the Excel data
json_template	Optional JSON string template — when provided, the response is structured to match the template schema

Returns a plain text string when json_template is omitted, or a structured dict when a template is provided.

---

Version history

Version	Changes
0.1.0	Initial release — CSV, DOCX, PDF, Excel, MarkItDown extractors