Jaxon is unlocking the power of AI for every corner of the organization. With rapid prototyping as our core focus, users train custom AI models and quickly iterate to find the winners. Using AI itself, Jaxon automates bottlenecks around data prep and model training. With "just enough" human supervision, auto-labeling, and configurable AutoML, models go from hypothesis to production-ready in days vs. months.

Jaxon is one of the first to offer a 'coding optional' platform that allows analysts and data scientists to collaborate on model training, align requirements, and validate model performance iteratively. Supported by techniques such as weak supervision, transfer learning, and unsupervised data augmentation, Jaxon-trained models are built faster with a higher confidence in the "ground truth". In a recent benchmark, Jaxon-produced models had 33% less error, with over 90% less human time.

Guided Learning: All human-driven data labeling is driven by a cost-aware algorithm that interactively alternates between each of the below labeling modes, optimizing a balance between gathered information and man-hour cost.

Weak Supervision: Simple models and rules can be used to provide low(er) confidence labels to large swaths of unlabeled training data. The inherent tradeoff between quality and quantity (of training labels) can be mediated by noise-aware aggregation and training techniques.

Semi-Supervised Learning: With a small seed of human-provided training labels, semi-supervised techniques actively search out other unlabeled examples that exhibit similar characteristics. Similar to weak supervision, this introduces a lever for strategic tradeoff between label quality and quantity.

Transfer Learning: Transfer learning has driven much of the modern deep learning renaissance, especially including general-purpose computer vision and NLP. Jaxon extends this notion to fine-tuning models not just for end tasks, but also as domain or organization-specific models that support the subsequent development of suites of highly-customized task-specific models.

AutoML: The Data-Centric AI paradigm holds that most iterative improvement in an AI modeling project lies with manipulations and improvements to the training data. Jaxon embodies AutoML to quickly frame a strong-enough model that is still lightweight in order to support rapid iteration on the training data.