Bruna: A Real-Time Multimodal Voice Agent with Hybrid Reasoning

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Advanced, short

Summary

Bruna is a data-centric smart voice assistant, detailed in a paper presented at the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) in Salvador, Brazil, in April 2026. This system is powered by multiple Large Language Models and is specifically designed to support Stilingue and Blip products. Its architecture aims to provide an enriched conversational experience for users. A key feature of Bruna is its ability to deliver strategic insights in real-time, enhancing its utility beyond basic conversational functions. The paper, authored by Evandro Fonseca, spans pages 11-13 of Volume 2 of the conference proceedings, published by the Association for Computational Linguistics.

Key takeaway

For research scientists developing conversational AI, Bruna's architecture demonstrates how combining multiple Large Language Models can create a data-centric voice assistant capable of delivering real-time strategic insights. You should consider this hybrid reasoning approach to enhance conversational experiences and provide tangible value in specific product ecosystems like Stilingue and Blip.

Key insights

Bruna is a real-time, multimodal voice agent using multiple LLMs for enriched conversations and strategic insights.

Principles

• Data-centric design enhances voice assistant capabilities.
• Multiple LLMs can power advanced conversational AI.

Method

Bruna's architecture integrates multiple Large Language Models to process multimodal input, enabling real-time conversational experiences and the delivery of strategic insights for Stilingue and Blip products.

In practice

• Integrate multiple LLMs for complex voice agents.
• Focus on real-time insight generation for business tools.

Topics

• Bruna
• Real-Time Voice Agent
• Multimodal Voice Agent
• Hybrid Reasoning
• Large Language Models

Best for: Research Scientist, AI Scientist, NLP Engineer, AI Engineer

Related on AIssential

• OpenAI brings GPT-5-class reasoning to real-time voice — and it changes what voice agents can actually orchestrate — OpenAI's new voice models modularize agent functions, reducing complexity and improving real-time conversational AI.
• Agent-based models for the evolution of morphological alternation patterns — Agent-based simulations, evaluated by an LLM-driven "AI Historical Linguist," model the emergence and spread of morphological alternations in language.
• Build an Emergency Helpline Voice Agent with LangChain — A low-latency, voice-only AI emergency agent can triage and dispatch services using concurrent STT, LLM, and TTS processing.
• Building real-time voice assistants with Amazon Nova Sonic compared to cascading architectures — Amazon Nova Sonic unifies speech processing for real-time, human-like voice AI, simplifying architecture and reducing latency.
• Why Emotion Matters More Than Sound — Building production-grade voice AI agents requires overcoming significant application-layer challenges beyond basic LLM integration.
• Grok Voice Think Fast 1.0: Build Voice AI Agents That Actually Think — Grok Voice Think Fast 1.0 integrates speech recognition, reasoning, and response into a single feedback loop for real-time, full-duplex voice AI.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.