Skip to content

Getting Started: Overview & Pathways

Welcome to the Episteme onboarding guide. Episteme is a research-grade platform designed to reconstruct theory graphs—multi-layered semantic, conceptual, and dialectical structures—from scientific literature.

Onboarding Pathways

Depending on your objective and background, choose one of two distinct pathways:

flowchart TD
    Start([Researcher / Reviewer]) --> Goal{Primary Objective}

    Goal -->|Explore Pre-computed Theory Graphs| PathStudio["Interactive Workbench (Episteme Studio)"]
    Goal -->|Extract Theory Graphs from Academic Prose| PathEngine["Computational Pipeline Engine"]

    subgraph StudioFlow ["Visual & Analytical Track (Zero Database Setup)"]
        PathStudio --> S1["uv run episteme-studio serve --demo"]
        S1 --> S2["Open http://127.0.0.1:8000"]
        S2 --> S3["Inspect L1–L3 Graphs, Dialectical Web & Tenability"]
    end

    subgraph EngineFlow ["Computational Track (Python & CLI)"]
        PathEngine --> P1["Install & Sync Dependencies (uv sync)"]
        P1 --> P2["Configure OpenAPI / LiteLLM Proxy & Neo4j 5.x (.env)"]
        P2 --> P3["Run Pipeline on Scientific Corpus (uv run python ...)"]
        P3 --> P4["Inspect Run Manifests & Query Neo4j Property Graph"]
    end

Pathway A: Interactive Workbench (Episteme Studio)

  • Target Audience: Epistemologists, domain researchers, and peer reviewers.
  • Key Advantage: Zero local database or API setup required.
  • Quick Launch: 
    uv run episteme-studio serve --demo
    Launches the visual workbench at http://127.0.0.1:8000 pre-loaded with demonstration runs, graph diffs, and metatheoretical tenability evaluations.
  • Learn More: Episteme Studio Workbench

Pathway B: Computational Pipeline Engine

  • Target Audience: Computational linguists, NLP researchers, and software engineers.
  • Key Advantage: Full programmatic control over extraction, model adapters, and graph projections.
  • Recommended Reference Stack:
  • Neo4j 5.x: Recommended graph store providing native property graph queries, vector indexing for TAG candidate retrieval, and GDS Leiden clustering.
  • Langfuse: Recommended platform for both Distributed Tracing (end-to-end token, cost, and latency visibility) and Remote Prompt Management (LangfusePromptProvider, enabling versioning, production labeling, and prompt iteration without modifying code).
  • Sequence:
  • Prerequisites: System requirements, Python 3.13+, and Neo4j 5.x.
  • Installation: Monorepo workspace setup via uv.
  • Configuration & Endpoints: Setting up OpenAPI-compatible endpoints and credentials.
  • First Run Tutorial: Executing pipeline_langfuse_full_run.py on authentic scientific texts.

The Three-Layer Epistemic Graph

Episteme does not produce flat, unstructured knowledge graphs. Instead, it extracts a tri-layer theoretical property graph:

Layer Visual Encoding Epistemic Function
Layer 1: Provenance Embedding Blue Nodes (Chunk, SourceDoc) Verifiable textual grounding, citation coordinates, and dense embeddings.
Layer 2: Empirical Ontology Green Nodes (Entity, Concept) Canonical entities, aliases, and factual relationships (Triple).
Layer 3: Dialectical TheoryNet Purple Nodes (TheoryAtom, Claim) Higher-level theoretical claims, formal SUPPORT/ATTACK vectors, and tenability scores.

Authentic Scientific Corpus

All tutorials and runnable examples in Episteme use authentic, peer-reviewed scientific literature located in packages/episteme-pipeline/examples/text/:

  • teachers_expectancies.md: Rosenthal & Jacobson (1966) Teacher Expectancies: Determinants of Pupils' IQ Gains (educational psychology).
  • planwirtschaft.md: Classical socialist calculation and market mechanism debate (economic philosophy, German).

Next Steps