Prompt-Based System Design
Exploring system design tradeoffs in a collaborative, reusable environment is the way!
Jaxon is an AI builder that guides developers through the design and implementation of AI systems.
Complex problems require systems of multiple models working together. Model systems designed on whiteboards and one-off digital diagrams are not reusable, don’t draw from previous system performance, and cannot recreate the merit of the design. Moreover, the knowledge goes with the designers when they leave.
System Design Tradeoffs
Software architects have to consider the -ilities: scalability, reliability, testability, etc. Adding machine learning to the equation only creates more -ilities and tradeoffs.
Robustness vs Complexity
If the task changes in the future, how hard is it to accommodate that? Is this worth the drawbacks of training and maintaining more models in tandem?
Data Frugality vs Effectiveness
Is the cost of acquiring more training data for a particular model within a system worth the overall inference boost?
Transparency vs Efficiency
Breaking a system down into discrete components makes it understandable, testable, and explainable, but this comes at the cost of breaking up a process that could train and infer as a single cohesive unit.
Jaxon creates the blueprints and then guides users building custom AI
Explore Design Tradeoffs
Prevent Unnecessary Work
Identify the Highest Impact Tasks
Formalize the Design Process
How does Jaxon design systems?
Jaxon presents design candidates to a user with equivalencies and tradeoffs, based on the initial, most basic version of the solution.
Visual transformations represent an understandable, structured way to alter a system design, while retaining overall functionality. These transformations, as well as design components, can be reused in future projects.
Jaxon includes a repository of system design knowledge that minimizes time and effort spent agonizing over small design decisions.
Jaxon embodies a formalism for designing and refining ML system architectures.
Visual, customizable transformations automatically update an architectural design, allowing before and after comparisons. These discrete transformations provide a trail of breadcrumbs so the design process itself can be captured—both the paths taken and those not taken.
Machine learning models, and the training data that drives them, are tuned to solve a specific task. A primary focus for Jaxon is ensuring that the tasks are right—e.g. the context of the overall system—before investing time and compute power into annotating data and training models.
System Design Complexity
Jaxon provides a collaborative canvas and toolbox to augment and formalize the ML design process. Combine the right foundational tasks in the right way to solve your complex data inference challenges.
