FindOrg — Named Entity Recognition for Organizations using LLMs

FindOrg is a Python package that performs Named Entity Recognition (NER) targeting organizations within a given text. It leverages large language models from OpenAI or Google Gemini to extract organization names and return them as a tidy pandas DataFrame.

Installation

pip install FindOrg

To enable saving results as Excel files (.xlsx), install with the optional extra:

pip install FindOrg[excel]

Quick start

from findorg import org

text = (
    "In the heart of Silicon Valley, a collaboration has emerged between "
    "global tech giants such as Google, Apple, and Meta, aiming to "
    "revolutionize the digital landscape."
)

# OpenAI (default provider)
df = org(text, api_key="YOUR_OPENAI_API_KEY")
print(df)

  Organizations
0        Google
1         Apple
2          Meta

Using Google Gemini instead

df = org(text, provider="gemini", api_key="YOUR_GEMINI_API_KEY")

API keys via environment variables (recommended)

If api_key is omitted, FindOrg reads it from the environment: OPENAI_API_KEY for OpenAI or GEMINI_API_KEY for Gemini. This keeps keys out of your source code:

df = org(text)                      # uses OPENAI_API_KEY
df = org(text, provider="gemini")   # uses GEMINI_API_KEY

Analyzing multiple texts at once

Pass a list of strings. The result gains a Text_ID column (1-based index of the input text):

texts = [
    "Petrobras signed an agreement with the World Bank.",
    "The event took place in Brasília, with no companies involved.",
]

df = org(texts, verbose=True)
print(df)

   Text_ID Organizations
0        1     Petrobras
1        1    World Bank

Saving results

df = org(text, save_path="organizations.xlsx")   # Excel (requires openpyxl)
df = org(text, save_path="organizations.csv")    # CSV

Arguments

Returns

pandas.DataFrame with one row per extracted organization:

• Single text input → column Organizations.
• List input → columns Text_ID and Organizations.

Extraction rules

• Geographic locations (countries, states, cities) are not extracted — organization names only.
• When an organization's name is followed by its acronym in brackets, only the full name is extracted.
• Texts with no organizations return an empty DataFrame.

Migrating from version 0.x

Version 1.0.0 introduces breaking changes:

The change of default model was required because OpenAI is shutting down gpt-3.5-turbo in October 2026.

How to cite

Neves, L. F. F. (2026). FindOrg: Named Entity Recognition for Organizations using LLMs (Version 1.0.0) [Python package]. https://pypi.org/project/FindOrg/

Contact

luiz.felipe@ufg.br