04Open source
Public on GitHub
Case-LLM
A local-first GraphRAG system that checks its answers against the graph.
- stores: graph for precision, vectors for nuance
- 2
- data leaving the machine
- 0
- acceptance scenarios, from logic to hallucination
- 5
Case-LLM turns unstructured podcast audio into a queryable knowledge graph, entirely on local models.
Whisper transcribes, an LLM extracts entities, Neo4j stores the relationships and ChromaDB handles semantic recall. Before it answers, it verifies that the path actually exists in the graph.
What is in it
- 01Hybrid retrieval: Cypher graph traversal combined with vector search.
- 02Intent routing between relational and semantic questions.
- 03Graph-path validation: if the relationship is not in the graph, it does not answer.
- 04Runs offline on Ollama and Whisper, with speaker diarization.