How Academic-Centric AI Projects are Driving Legal Tech R&D

When organizations lack the resources to rigorously test cutting-edge AI tools, who fills the research gap? A collaboration between Worcester Polytechnic Institute (WPI), HaystackID®, and Relativity® demonstrates how academic partnerships can accelerate innovation. A team of WPI undergraduates conducted comprehensive benchmarks of large language models (LLMs) and developed a working prototype for AI-driven legal document classification. The nine-month project tackled real challenges facing eDiscovery professionals: how do different LLMs perform on legal datasets, what processing methods yield the highest accuracy, and how can confidence thresholds improve classification reliability? The students evaluated over 10 models, introduced novel enhancements to hypergraph neural networks, and built a live web application that achieved 94% accuracy with Claude 3.7 Sonnet—surpassing many industry benchmarks. This webcast will explore the practical insights emerging from academic-industry collaboration, including zero-shot prompting techniques and document length optimization, as well as the finding that documents without unique keywords often classify more accurately than those with specialized legal terminology. Beyond the technical results, the project exemplifies a scalable model for legal tech R&D that benefits both industry partners seeking rigorous testing and academic institutions preparing the next generation of AI-literate legal professionals. Expert Panelists  + Aron Ahmadia, PhD Senior Director, Applied Science at Relativity + Roee Shraga, PhD Assistant Professor, Computer Science, WPI + John Brewer [Moderator] Chief Artificial Intelligence Officer and Chief Data Scientist, HaystackID Source: HaystackID