T-Episteme: Theory Graph Language Pipeline¶
Welcome to the documentation for grund.Episteme, a cutting-edge pipeline for constructing theory graphs from scientific literature.
What is T-Episteme?¶
grund.Episteme is a research library designed to construct theory graphs—structured representations of theoretical frameworks—from academic texts in German and English. While designed for computational epistemologists, NLP researchers, and philosophers of science, the pipeline itself focuses on an agnostic and approach for knowledge graph construction and may serve different use cases as well.
Our pipeline abstracts unstructured scholarly documents into queryable semantic structures. Unlike traditional knowledge graphs that focus on factual relationships, theory graphs capture the conceptual and argumentative structures that underpin scientific theories. The pipeline transforms unstructured scholarly documents into a property graph, that can be stored in any graph database of your choosing. The episteme-pipeline produced a three-layer architecture:
- Layer 1 (Provenance Embedding): Original documents embedded and queryable in the graph.
- Layer 2 (Knowledge Graph): Abstraction from original sources as a knowledge graph.
- Layer 3 (Theory Framework): Higher-level theoretical constructs and argumentative structures.
Documentation Pillars¶
The documentation is organized into six foundational pillars designed to serve researchers, system architects, and software engineers:
-
Getting Started & Tooling
Zero-friction onboarding, environment configuration, step-by-step first run, and the interactive Episteme Studio workbench.
-
Research & Epistemology
Formal graph models (\(\mathcal{G}_{\text{TheoryNet}}\)), structuralist philosophy of science (Wissenschaftstheorie), and the PNAS Nexus publication.
-
Architecture & Decisions
System-wide engineering trade-offs, dual-store graph projection, cache invalidation state machines, and 18 Architecture Decision Records (ADRs).
-
Pipeline & Engineering
Production-grade, deterministic 5-phase extraction pipeline, theoretical enrichment post-processing, and distributed Langfuse tracing.
-
Epistemetrics & Validation
Standalone quantitative evaluation library measuring empirical creativity, theoretical unification, tacking-paradox homogeneity, and Lakatosian dynamics.
-
Reference & Contracts
Strict Pydantic phase I/O boundaries, Neo4j Cypher property graph schemas, Python APIs, and the unified bilingual conceptual glossary.
System Architecture & Processing Flow¶
