Session: 2 for 1: Accelerating the next generation of multimedia with GPUs / From Retrieval to Real-Time Action: RAG, ReAct, and Flow Orchestrator in Building Smarter AI Bots
Joshua Alphonse: Accelerating the next generation of multimedia with GPUs
This session will explore how to create a next-generation video broadcasting/live stream service! In our ever-changing world of multimedia, performance and compatibility with AI are what will make your service stand out. Traditionally, we would use FFmpeg to process videos. The issue is that there is no multi-language support, and it’s slow!
In this talk, I’ll introduce the Babit Multimedia Framework (BMF) and how it simplifies the process of building a live broadcasting service. With a focus on dynamic video source handling, layout control, audio mixing, and GPU acceleration, we’ll explore how to use BMF to quickly deploy robust broadcasting solutions. We’ll highlight BMF’s modularity and flexibility, making it an indispensable tool for developers seeking to break into the live streaming market or optimize their current solutions.
Here are some key takeaways:
- An in-depth look at how BMF addresses the complexities of building live broadcast services by providing a straightforward high-level API.
- Insights into the customizable video pipeline that supports dynamic source input and layout management for a professional broadcasting experience.
- Demonstrations of how BMF facilitates multi-language development and allows for rapid adjustments to the streaming pipeline.
- A walkthrough of how to create a streaming pipeline to an RTMP server, revealing the practical implications and real-world applications.
- Best practices for utilizing CPU+GPU acceleration for high-performance video processing during live broadcasts.
Aria Chang: From Retrieval to Real-Time Action: RAG, ReAct, and Flow Orchestrator in Building Smarter AI Bots
You’ve likely heard of large language models (LLMs) and Retrieval-Augmented Generation (RAG), but what about ReAct? Join Microsoft Software Engineer Aria Chang as she explores these cutting-edge AI techniques and introduces ReAct—a unique method that enables LLMs to reason and act dynamically, responding to context and adapting in real-time.
In this session, Aria will showcase Flow Orchestrator, an experimental tool from Microsoft’s Semantic Kernel that implements ReAct to orchestrate AI workflows seamlessly. Aria will walk through a mock coding interviewer bot application built using Flow Orchestrator, showing how ReAct allows it to retrieve relevant information, make decisions, take actions, and engage in contextual dialogue—essentially conducting a human-like interview.
For AI enthusiasts, developers, and those intrigued by next-gen conversational bots, this talk provides the practical tools and insights to build applications that don’t just retrieve information—they think on their feet. Step into the world of responsive AI, where conversation feels more authentic than ever.
This session will be recorded
