Assistants & RAG
Assistants that answer from your knowledge, with sources, tested before every release.
A knowledge assistant, end to end
Ingestion keeps the index current; each question is checked for scope, answered with sources and tested before release.
Point at a component to light its links; select it for what it does here and the practice that makes it right on AWS.
Generative AI we build and run
NGT's GenAI service lines. Each ships with tests, tracing and a hand-over to people.
How a RAG service is built
The steps NGT follows for a knowledge assistant, from source system to tested answer.
Collect
The owning system publishes document metadata to a queue; a consumer picks it up.
- For example, the owning system publishes to a message broker such as ActiveMQ or Kafka; a small consumer service feeds the orchestrator.
- Each message is checked against a schema, and processing time is logged against an SLA.
From demo to production
Most demos work. These differences decide whether one still works with real users.
Questions this raises
The questions this part raises, and their answers.
Usually not first. Most enterprise questions are answered by good retrieval over well-prepared documents.
We fine-tune for a fixed format or tone, or when a smaller model must do a larger one's job for cost or data-location reasons.