semantica

Project Overview

Semantica is graph-native AI knowledge infrastructure. It ingests fragmented enterprise data, extracts entities and relations, builds a Context Graph and knowledge graph, and runs graph analytics and causal reasoning on top. Graph construction, reasoning, and decision provenance are deterministic and do not require an LLM; when an LLM is needed, every major provider can be swapped in. It targets regulated domains such as finance and healthcare that demand explainable and auditable AI.

Replace plain vector similarity with a structured semantic graph, so every AI conclusion can be traced to its source.

Key Features

• Data ingestion: connect multi-source enterprise data and extract entities, relations, and attributes
• Context Graph: organize business context into a structured, queryable semantic graph
• Deterministic reasoning: graph construction and inference do not depend on an LLM, so results are reproducible
• Ontology management: define entity meanings and rules with OWL, SHACL, and SKOS standards
• Decision provenance: every conclusion carries a full audit chain to answer why
• Polyglot storage: supports both RDF and property graphs across multiple graph databases
• Optional LLM: plug in OpenAI, Anthropic, Gemini, and more via semantica.llms, vendor neutral

Target Audience

• Data engineers building enterprise knowledge graphs
• Agent teams that need explainable reasoning
• AI leads in regulated industries such as finance and healthcare
• Developers building GraphRAG retrieval systems

Quick Start

1. Install:

pip install semantica

2. Configure data sources and ontologies per the official docs to start ingesting and building graphs

Core Advantages

• Explainable: structured graphs and provenance chains make conclusions auditable
• Deterministic: inference does not rely on probabilistic models; same input, same output
• No vendor lock-in: open source, optional LLMs, and open standards

System Requirements

Notes

• Scope of explanation: it explains system-level reasoning chains and does not expose what happens inside an LLM
• LLM optional: graph construction and provenance never require an LLM; connect one only when needed