Here is your today’s AI Dev Brief from Marktechpost, covering core research, models, infrastructure tools, and applied updates for AI developers and researchers.
NVIDIA Releases PersonaPlex-7B-v1: A Real-Time Speech-to-Speech Model Designed for Natural and Full-Duplex Conversations
NVIDIA Researchers release PersonaPlex-7B-v1, a full duplex speech to speech model that replaces the usual ASR to LLM to TTS pipeline with a single dual stream Transformer. The system listens and speaks at the same time using Mimi encoders and decoders at 24 kHz and generates text and audio tokens jointly for fast turn taking, interruptions, and natural backchannels. Persona control is handled by a voice prompt that sets timbre and style and a text plus system prompt that defines role and business context. Training combines more than 1,200 hours of Fisher conversations with about 2,200 hours of synthetic assistant and customer service dialogs. On FullDuplexBench and ServiceDuplexBench, PersonaPlex reaches high takeover rates with sub second latency.....… Read the full analysis/article here.
Black Forest Labs Releases FLUX.2 [klein]: Compact Flow Models for Interactive Visual Intelligence
Black Forest Labs releases FLUX.2 [klein], a compact rectified flow image model family that targets interactive visual intelligence on consumer hardware. The series includes 4B and 9B variants that support text to image, single image editing, and multi reference generation in one architecture. The distilled models run with 4 sampling steps and reach sub second latency on a single modern GPU, while base models use longer schedules for fine tuning and research. Quantized FP8 and NVFP4 versions, built with NVIDIA, provide up to 1.6 times speedup and about 40 percent lower VRAM for FP8, and up to 2.7 times speedup and about 55 percent lower VRAM for NVFP4 on RTX GPUs. With Apache 2.0 licensing for 4B and open weights along with broad ecosystem support, FLUX.2 [klein] is ready for real time visual tools and agent workflows....… Read the full analysis/article here.
Project Notebooks/Tutorials
▶ [Open Source] Rogue: An Open-Source AI Agent Evaluator worth trying Codes & Examples
▶ How to Build a Self-Evaluating Agentic AI System with LlamaIndex and OpenAI Using Retrieval, Tool Use, and Automated Quality Checks Codes Tutorial
▶ How to Build a Safe, Autonomous Prior Authorization Agent for Healthcare Revenue Cycle Management with Human-in-the-Loop Controls Codes Tutorial