Jaxon’s Domain-Specific AI Language (DSAIL)

Drastically reduce or eliminate hallucinations:

dsail z graphic

Rigorously automate 

the design cycle

dsail eye graphic

Reason about the design

in a very abstract way

dsail cloud chart graphic

Prototype with meta

models and simulated data

Your secret weapon

DSAIL formalizes the process of designing complex AI systems. It organizes various elements such as requirements, abstract concepts, metadata, and overarching frameworks. By simulating model performance, users are able to explore design trade-offs and get the design right the first time. 

 

DSAIL helps developers

DSAIL uses a patent-pending approach to specify, consider, reason about, and decide on the space of what is possible, before ultimately choosing a feasible (optimal or near-optimal) solution – and before designing the implementation that codes the how. Breaking things down to its core, DSAIL allows a user to:

  • Specify in abstract terms the problem, or task, to be solved.
  • Do so with declarative natural and/or formal languages to state unambiguously what is needed for a performant task, but not how to accomplish the task.
  • Do that by asserting certain constraints on any candidate solution that must simultaneously hold true for it to be a feasible solution.

All properties and relationships are represented in a graph that serves as a ‘Model of Computation’, enabling users to:

The Image Description Fact Checker

DSAIL’s ‘Fact Checker’ is at the heart of addressing the hallucination problem. By defining the appropriate categories, functors, and natural transformations, DSAIL creates a formal system that aligns with your goals of optimizing user-chosen aspects through graph rewriting.

  • Explore the design space of feasible solutions as abstractly as practically possible to defer deciding on implementation details as late as possible.
  • Use decision aids of formal models served by machine executable analysis, search, and validation.
  • Assess the degree to which any feasible solution is optimal or near-optimal.
  • And ultimately finalize the architecture, auto-generate the code, and verify the task solution.

Eager to learn more?

Download the White Paper