climatextract 0.3.2

Extract CO2 emissions data from PDF sustainability reports using LLMs

climatextract

📖 Documentation: gist-sustainability.github.io/climatextract

climatextract is a retrieval-augmented generation (RAG) pipeline that surfaces CO₂ emissions data from corporate sustainability reports. It embeds PDF pages, ranks relevant context, and prompts a large language model to extract Scope 1-3 emissions into structured tables for downstream analysis.

Background

This project began as the team's submission for the 2024 ClimateNLP workshop at ACL. Built by the LMU SODA Lab in collaboration with the Data Service Centre of Deutsche Bundesbank, climatextract combines research around ESG reporting and Intelligent Document Processing to automate what was previously a tedious manual annotation process.

This repository is organized as follows:

• climatextract: package source code
• data: source data to be analyzed
• docs: package documentation (built with mkdocs)
• tests: acceptance tests

Setup

Python environment

It is recommended to run the code in a virtual environment using at least Python 3.11.

First, check out the code, then create a virtual environment and install all dependencies:

cd climatextract
python -m venv co2_info_extraction
source co2_info_extraction/bin/activate
pip install -e .

See the Installation guide for additional steps and alternative deployment options you have.

Usage

Place your PDF sustainability reports in the data/pdfs/ directory, then run the extraction pipeline:

from climatextract import extract

result_path = extract("./data/pdfs/company_2023_report.pdf")

Results are saved as CSV files in output/<run-id>/. See the Quickstart for more examples.

Configuration

Extraction behavior is controlled via a climatextract.toml file in your working directory. It lets you configure the LLM model, embedding model, prompt type, year range, semantic search parameters, and more. See the Configuration guide for all available options.

Running tests

python -m pytest

See tests/README.md for details on the acceptance test suite.

Documentation

The full documentation is hosted at gist-sustainability.github.io/climatextract and covers usage, configuration, architecture, and API reference:

Section	Description
Installation	Detailed setup instructions
Quickstart	First extraction walkthrough
Configuration	All TOML configuration options
Custom Providers	Plug in a non-Azure LLM or embedding backend
Architecture	Pipeline design and components
Prompts	How extraction prompts work
Evaluation	Measuring extraction quality
API Reference	Public API functions

To build and serve the docs locally:

pip install -e '.[docs]'
mkdocs serve