We publish our work on neural retrieval, evaluation, and AI-native search. Our research directly powers the products we ship — and we believe in sharing it with the broader community.
The problems we are focused on
Advancing the state of the art in dense retrieval and re-ranking, with a focus on semantic understanding and real-world query diversity.
Extracting structured data (companies, people, events) from unstructured web content at scale, with high precision and recall.
Building rigorous evaluation frameworks that measure what actually matters for AI applications, not just traditional IR metrics.
Pushing the limits of what is possible at low latency and global scale, from index construction to serving billions of queries.
Selected papers from the RunoxAI research team
Priya Mehta, Jordan Kim, Alex Rivera
We present a novel approach to web-scale neural retrieval that achieves state-of-the-art performance on standard benchmarks while maintaining sub-200ms latency. Our method combines dense retrieval with a proprietary re-ranking stage trained on billions of real-world queries.
Priya Mehta, Sarah Williams
Existing search evaluation frameworks focus on keyword-based retrieval. We introduce WebBench, a benchmark suite specifically designed to measure the quality of semantic and neural web search, covering diverse query types, freshness requirements, and domain-specific tasks.
Jordan Kim, Priya Mehta
We study the problem of extracting the most relevant text passages from web pages given a user query, with the explicit goal of minimizing token consumption in downstream LLM applications. Our approach reduces token usage by up to 75% while maintaining answer quality.
Sarah Williams, Marcus Chen
Stale cached data is a major limitation for AI applications that require up-to-date information. We describe LiveCrawl, a system that fetches and processes web pages in real time, enabling AI agents to access the most current information available on the web.
Priya Mehta, Alex Rivera, David Park
We present a system for retrieving structured company information from the web, integrating entity recognition, cross-document inference, and confidence estimation. Our approach powers RunoxAI's company search vertical, which serves thousands of enterprise customers.
We are looking for researchers and engineers who want to work on foundational problems in information retrieval and AI. Come build with us.