Leibniz Lab - YAGO and PyReason Assessment

This project combines a YAGO knowledge graph subset, LLM-generated grounding facts, and PyReason inference.
The goal is to show that explicit rules can fire on top of trusted graph facts plus new facts suggested by an LLM.

Project Overview

• Knowledge graph input: src/data/yago_subset.graphml
• Rules input: src/data/rules.json (five PyReason-compatible rules)
• LLM grounding output: src/output/llm_facts_pyreason.json
• Inference output: src/output/inference_result.json and src/output/inference_rule_trace_edges.csv
• Human summary: src/FINDINGS.txt

The dashboard overlays three layers in one graph view:

• YAGO edges (solid)
• LLM fact edges (dashed)
• Inferred rule-head edges (thick)

Quick Start

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env

Set OPENAI_API_KEY in .env before running inference.

Data and Inference Flow

1) Build the YAGO GraphML subset

python -m src.buildYagoSubsetFromHuggingFace

How this works: buildYagoSubsetFromHuggingFace streams VLyB/YAGO3-10 and keeps a small fixed set of required triples.
It writes only that minimal subset to src/data/yago_subset.graphml.

2) Run the full pipeline (LLM + PyReason)

python -m src.runInference

This step:

• reads yago_subset.graphml and rules.json
• calls OpenAI once per showable rule
• converts model output into PyReason fact JSON
• runs reason()
• writes inference artifacts under src/output/

Dashboard

Run:

./run_dashboard.sh

The dashboard is read-only and loads files from disk.

• Default mode: reads src/output/
• Demo mode: reads frozen sample artifacts in src/brunoOutput/

Demo mode command:

LEIBNIZ_USE_BRUNO_OUTPUT=1 ./run_dashboard.sh

runInference always writes to src/output only.
It does not modify src/brunoOutput.

Reset Generated Outputs

python -m src.cleanOutputs

This removes generated files under src/output and keeps:

• src/data/rules.json
• src/data/yago_subset.graphml

Useful Output Files

• LLM prompt/reply audit: src/output/llm_openai_prompts_and_replies.json
• Merged LLM payload: src/output/llm_full_payload.json
• PyReason fact file: src/output/llm_facts_pyreason.json
• Inference bundle: src/output/inference_result.json
• Full trace CSV: src/output/inference_rule_trace_edges.csv

macOS SSL Note

If OpenAI calls fail with CERTIFICATE_VERIFY_FAILED, install dependencies again and retry:

pip install -r requirements.txt
python -m src.runInference