architecture
How our composite AI architecture works
GPT For Numbers is the AI architecture designed to support Decision Intelligence processes in highly regulated sectors and to ensure maximum protection against errors and hallucinations.
Human language understanding
A language model fine-tuned to understand what you are asking and extract the business entities from your question, that will be used to retrieve the correct data.
Business Knowledge Graph
A network of interconnected information tasked with extracting metadata necessary to transform data into conversations. The BKG establishes semantic connections, evolves based on requests, and ensures results are always accurate and contextual.
Orchestrator
A collection of Expert Models
GPT For Numbers is based on the Small & Wide Language Models methodology, which combines a collection of Expert Models of varying sizes, with each node being a model trained to perform a specific task. This is all orchestrated by a central control system called the AI Orchestrator, which implements the necessary AI Guardrailing to reduce the risk of errors and hallucinations.
Insight generation
A model that reveals hidden insights from business data and conducts automatic analyses without any technical input.
Anomaly detection
A model that actively monitors business data and sends notifications about anomalies and events of interest.
Smart suggestions
A model that identifies the best next question, providing detailed guidance for further, deeper analysis.
LLM for generative answers
It generates answers for non-technical people, with unique insights and follow-up steps.
Dynamic Data Representation Engine
It creates interactive and business-friendly data visualizations, based on users' needs and requests.
