RinggAI/ringg-router-e2b

Overview

Ringg Router E2B is a specialized 5.1 billion parameter decision model, fine-tuned by Ringg AI from google/gemma-4-E2B-it. Its core function is to quickly process short conversations and a list of options, then output a JSON object indicating the best option, optionally extracted values, and a one-sentence rationale. This model is particularly optimized for multilingual Indian phone conversations, handling English, Hindi, Hinglish, and other code-mixed speech, as well as Bengali, Telugu, Tamil, Kannada, Malayalam, Marathi, and Gujarati.

Key Capabilities

• Rapid Decision-Making: Designed for fast routing and intent decisions within voice or chat agents, serving as an IVR replacement or for support triage.
• Structured Extraction: Capable of extracting named fields from short conversations, including those in Indian languages and code-mixed text.
• Tool/Function Selection: Facilitates tool or function selection, including scenarios where no tool applies.
• Multilingual Support: Extensive training on public and proprietary datasets for Indian languages, with oversampling for underrepresented languages like Telugu, Kannada, and Gujarati.

Performance Highlights

Evaluations on held-out and validation splits demonstrate significant improvements over its base models. For instance, it achieves 98.9% accuracy on intent routing and 99.6% on tool/function selection on the held-out split. It also shows strong performance in entity and slot extraction, with 85.6% field accuracy and 62.7% all fields correct for entity extraction on Indic and multilingual datasets, substantially outperforming Gemma-4-E4B-it.

When to Use This Model

Ringg Router E2B is ideal for applications requiring quick, accurate, and structured decisions from conversational input, especially in multilingual environments involving Indian languages. Its small size and speed make it suitable for real-time voice agent interactions, where low latency and high precision in routing and data extraction are critical.