KIBA-D code Reference

LLM-based information extraction and evaluation framework.

Imports config on package initialisation to set global constants and load environment variables.

Modules:

Name	Description
config	Package-wide constants and environment variable loading.
predict	Entrypoint for running LLM inference on preprocessed documents.
evaluate	Entrypoint for evaluating predictions against reference datasets.
preprocessing	PDF-to-markdown conversion utilities.
metric	Base class defining the metric interface.
data_integration	Utilities for downloading and syncing source data (Zotero, Nextcloud, database).
dataset	Dataset I/O helpers for CSV, JSON, and prediction files.
extractors	Structured-output extractors for parsing LLM responses.
llms	LLM provider integrations (OpenAI, vLLM).
metrics	Concrete metric implementations (e.g., F1, confusion matrix, error collection).
schema	Shared type definitions and schema utilities.
hydra_callbacks	Hydra lifecycle callbacks.
utils	General-purpose helpers.