Approaches like API RAG (retrieval-augmented generation) provide data for more accurate and reliable AI development. Baseline RAG uses a vector database (vector embeddings) to provide context for model’s responses. Graph RAG augments RAG with knowledge graphs to use data relationships.
Smabbler's API RAG (Graph RAG as-a-Service) utilizes Galaxia graph knowledge model to simplify and shorten RAG pipeline. Galaxia automates knowledge graph creation, context enrichment, and retrieval via simple API. There is no need for graph or vector databases. And the whole thing runs on CPUs which means lower operating cost and improved scalability.
Structured and contextual knowledge integration
Improved accuracy and reduced hallucinations
Enhanced relevance and precision of information retrieval
Better handling of complex queries and relationships
Scalability and flexibility for domain-specific applications
More efficient use of memory and computation
Continuous updating and dynamic knowledge integration
Enhanced explainability and transparency
Smabbler's API RAG is a few clicks flow with a text data body as an input and retrieved text as an output.
LOAD – load your text data source.
BUILD – use BUILD LANGUAGE GRAPH function.
ACTIVATE – activate your GRAPH MODEL.
Your API RAG is ready to work with an LLM of your choice.
PARSING - sentence / text parsing to form intermediate representation - tokenization, lemmatization, part of speech (POS) tagging, dependency parsing. The text is represented as a graph structure.
ENRICHMENT - enriching text with semantics, similarities, definitions, context and relations. The enriched text is represented as a highly interconnected graph structure, creating a user’s Language Graph.
STORAGE - Language Graph (Graph Load) in the form of a file (JSON) is stored in the database in the user's account (in Smabbler Portal).
ACTIVATION - during the activation (ACTIVATE) graph load connects to the Galaxia graph language model, to create a user Graph Model.
PARSING - similarly to the Language Graph creation process, the user query is parsing to form intermediate representation.
ENRICHMENT – the user query is enriched with semantics, similarities, definitions, context and relations.
MATCH - the graphed and enriched query is matched to the Language Graph to provide most relevant results.
[1] KB (Knowledge Base) - here, a collection of information resulting from the transformation of raw text into structured data, extended with similarities, context, and relationships.
[2] Language Graph (or Graph Load) - a graph created by transforming a text body into a graph structure, extended with knowledge and lexical layers to add relations, semantics, and context.
[3] Graph Model (or Graph Language Model) - a language graph with embedded NLP-based retrieval algorithms that analyzes text using itself as a knowledge base.