I’m sharing Semantica, an open-source framework for building knowledge-driven RAG systems.

Most RAG pipelines rely on vector similarity over text chunks. This works well for simple retrieval, but breaks down when systems need:

explicit relationships

multi-hop reasoning

global context

explainability

Semantica takes an ontology-first approach:

Raw data is transformed into entities and relationships

Knowledge is stored as a knowledge graph

Retrieval combines graph traversal and vector search

LLMs operate on structured knowledge, not just text

The goal is to make RAG systems more reliable and traceable, especially for domains like healthcare, enterprise search, and research.

Semantica includes:

Semantic layers

Knowledge engineering pipelines

RDF/OWL ontology ingestion

Knowledge graph construction

GraphRAG-style retrieval

GitHub: https://github.com/Hawksight-AI/semantica

The project recently crossed ~300 stars.

I’d appreciate feedback from people working on RAG, IR, or knowledge graphs.

Thanks.