Build Knowledge Graphs with txt2kg

Why it matters

Graph retrieval is only as good as the triples behind it. This playbook runs the whole path, from uploaded text to a graph-backed answer, on one machine.

What to practise

The playbook targets DGX Spark and DGX Station. Start the stack with ./start.sh — ArangoDB with Ollama by default, or Neo4j with --neo4j or --vllm — and choose the extraction model. Upload Markdown, text or CSV documents, and watch them chunked into subject–predicate–object triples. Explore the graph in 2D or 3D, then ask questions that graph traversal answers. Compare triple quality across model sizes.

How it fits

The playbook’s steps go from extraction straight to storage, so note where each of the checks 1.7.2 separates — type checks, entity resolution and source verification — would go. Ask whether its entities keep the identity and source that 1.7.1 requires, and how far 1.7.3 would let a traversal expand.