ej

ej is engineered for targeted decision-making, providing probability distributions for various questions based on a given state. Key features include: • Returns probability distributions for decision questions • Handles choice, yes/no, and ordered scoring • Operates efficiently on CPU without GPU requirements • Packaged as a single, compact 11.4 MB model file • Includes tools for training, evaluation, and adaptation This compact decision model is ideal for scenarios requiring rapid classification and routing without the overhead of large language models. It runs entirely offline within your service, eliminating API calls, GPU rentals, and complex JSON parsing. Every question about a record is answered in a single pass, providing a complete probability distribution for each option. This allows teams to automate straightforward decisions and escalate more complex cases based on defined probability thresholds. ej is designed for engineering teams that need to integrate decision-making capabilities directly into their applications. It's perfect for use cases like ticket routing, checking tool calls, and scoring urgency, all with precise probability outputs. The open weights and Apache-2.0 licensed code ensure flexibility and transparency for developers.