CUSTOMIZE

Creation is simple: Give your new AI model a name.

Our system gives complete control and flexibility for defining every aspect of an encoder-decoder transformer from scratch via hyperparameters. Our default models have around 60M parameters, establishing a powerful SLM (small language model) by default for modern day tasks. 

Platform capabilites surround model creation, maintenance, and usage. Control access, visibility, permissions, and every aspect of your business in our completely vertical AI platform. Audit trails, approval mechanisms and data encryption make this AI platform flexible to fit any security needs.

Our platform enables a rich and customizable combination of:

• Knowledge graphs

• KG-Trie constraints

• Human inspired bio-mimicry learning patterns

Reason's AI training regiment boasts a 90% reduction in the amount of training data needed. This simple approach makes AI accessible to those with limited amounts of data, who may not have the resources or time to create AI models of their own.

EVALUATE

Evaluation is simple: Give an input, receive an output.

Our platform's models are highly specialized and focused on a clear task. 

This is not a conversation like a GPT/chatbot. This is not statistical word matching like a decoder-only GPT that are limited by masked self-attention and unidirectional causality.

Our AI understands context by leveraging bidirectional self-attention, masked self-attention and cross-attention. In GPT, cross-attention is mathematically absent.