Use Cases
While the genesis of TrustGraph
was to evolve GraphRAG
by aligning with open source technologies and standards like RDF
, the need for a simple way to deploy AI infrastructure became apparent. In addition to simplfying AI deployments, the AI Engine
aproach of TrustGraph
has three key use cases: data enhancement, customization and extensibility, and exclusive deployments.
Data Enhancement​
TrustGraph
can ingest large amounts of unstructured data scattered across documents and structure that knowledge as RDF
knowledge graphs and mapped vector embeddings. This structuring process uncovers patterns and relationships that would previously require manual, human analysis to discover. Sample uses:
- Regulatory and Compliance Analysis
- Financial Report Analysis
- Academic Research
- Legal Research
- Legal Discovery
- Social Graph Analysis
- Message Logs Analysis
- Anomaly Detection
- Analysis Report Generation
Customization and Extensibility​
There are three concepts central to the customization and extensibility of TrustGraph
:
- Extraction Tailoring
- Reusable Knowledge Cores
- Apache Pulsar Pub/Sub backbone
Extraction Tailoring​
The default extraction process in TrustGraph
is a Naive Extraction
. A Naive Extraction
has no prior knowledge of the data to be ingested. The default TrustGraph
extraction is also designed to work with any LLM. This extraction process can easily be tailored by:
- Optimizing Prompt structure for a particular LLM
- Optimizing System Prompts for a particular LLM
- Defining important terms and concepts
- Definiting custom structure to enchance the RDF
Reusable Knowledge Cores​
The extracted knoweldge graph and mapped vector embeddings become a knowledge core
. Building knowledge cores
is a one-time process as they can be saved, shared, and reloaded. The concept of knowledge cores
enables loading only the knowledge needed for the AI Engine
. This approach also enables data access management.
Pub/Sub Backbone​
Apache Pulsar is an enterprise-grade pub/sub backbone that runs services connected by data processing queues and schemas. Processing modules easily integrate into TrustGraph
as either a consumer
, producer
, or consumer/producer
of data processing queues. New modules can subscribe and publish to existing queues or define their own.
Exclusive Deployments​
While TrustGraph
supports cloud-hosted models through Anthropic
, AWS Bedrock
, AzureAI
, AzureOpenAI
, Cohere
, Google AI Studio
, OpenAI
, and VertexAI
, many users want to maintain full control over their data. Sending sensitive data to external APIs opens the door to security weaknesses and in many cases, violates organizational policies. To solve this problem, TrustGraph
deploys a full end-to-end AI Engine
using Ollama
or Llamafile
. The Ollama
or Llamafile
deployments allow a Language Model to run in any environment, including a laptop (although running the models on a laptop is not advised because of thermal management issues). This approach provides a fully self-contained and secure
deployment for unlocking the power of GenAI
with open Small Language Models (SLMs).