AzureFeeds

As a Director of AI Product Marketing at Azure, I have spent the last three years deep in the GenAI ecosystem. From internal chatbots that evolved into multi-model agent orchestrations, I feel like I am witnessing history in my job and life. Every day I wonder what is coming next, to build on top of everything we have developed in just a short amount of time. Now I am excited to share the newest chapter in this incredible run of innovation with the announcement of Microsoft’s launch of NLWeb today, an open project aimed at facilitating the creation of natural language interfaces for websites, enabling users to interact with site content through natural language queries. The initiative strives to empower web publishers by making it easier to develop AI-driven applications that enhance user experience and site discoverability. 

NLWeb simplifies the development of natural language interfaces by utilizing existing semi-structured data formats and LLM tools, making it accessible for all web publishers.  This exciting technology was developed by R.V. Guha, the creator of RSS, RDF and Schema.org, who recently joined Microsoft as CVP and Technical Fellow.

As an open project, NLWeb has a growing list of other contributors from Microsoft and the open-source community including Inception. I had a chance to ask Inception about their early contributions to NLWeb as the GitHub repo is launched to the developer community.

Q1: What inspired your team to try NLWeb? 

At Inception, we are building a new generation of ultra-fast LLMs, which we envision will enable a new suite of real-time user experiences. We knew that large creator platforms needed better solutions to provide these experiences to their users. So, we were excited when Microsoft approached us asking us to integrate with NLWeb. From a technology perspective, we understand that delivering seamless user experiences requires fast and powerful LLMs that can perform efficient search, retrieval, and reasoning. These are exactly the use cases for our Mercury suite of models — our LLMs generate tokens in parallel using diffusion technology, which means they achieve ultra-low latencies while maintaining superior quality.

Q2: How did the setup process go? Any surprises?

The setup process was straightforward – the NLWeb team has developed a well-documented and modular GitHub repository that enabled us to easily integrate models hosted via our API. We were pleasantly surprised by the attention to detail throughout the code, such as the highly optimized asynchronous routines integrating the retrieval and generation capabilities. Given our own focus on latency-sensitive applications, we think that en-users will really benefit from this.

Q3: How are you blending NLWeb with your current experience?

We are working to offer our dLLMs through Azure AI Foundry. Accordingly, our customers will soon be able to use NLWeb fully on Azure, which will further reduce latency.

Q4: If NLWeb reaches its full potential, what could it unlock for your users or the web?

NLWeb will create a new kind of digital ecosystem that provides users with instant access to high-quality, hallucination-free content. At the same time, it creates a first-of-its-kind incentive structure for content creators and websites to embrace generative AI as a tool that brings them closer to their audiences. At Inception, we are committed to advancing generative AI breakthroughs that make such rich user experiences possible.