How to Leverage All of the World's Research for Scientific Discovery with GenAI

A leading MedTech company aimed to speed up its manual research process for hundreds of scientists while ensuring top accuracy. With Activeloop's sub-second, multi-modal AI search, they connected all of the world's research papers to LLMs, cutting research time from weeks to days, advancing medical device development and patient outcomes. Learn how.

November 13, 2024

Accelerating Medical Research with AI: A MedTech Company’s Story

Imagine doing scientific research at scale, having to sift through millions of research articles across PubMed, internal research notes, and patient data including MRIs, CT scans, and more.

The manual search process is slow, manual, and cross-referencing different, rapidly evolving data sources is prone to errors, ultimately delaying progress.

A leading MedTech company tackled this challenge in collaboration with Activeloop and Intel. Here's what they were able to achieve (watch a 1.5 min demo):

❝

“This end-to-end neural search capability by Activeloop has transformed how we make scientific discoveries,” said a senior research director. “We can now analyze connections instantly and surface insights that would have taken months to find manually.”

VP of Research

The goal was to connect internal data (patient reports, MRIs, CT scans, research notes) and external research (e.g., PubMed articles) to speed up scientific research and obtain fast, accurate answers to complex questions involving multi-modal data from different sources.

Deep Lake, combined with Intel’s 5th Gen Xeon® processors, provided an efficient way to search, connect research data, and integrate any data type with Large Language Models (LLMs) to enhance insights. Tasks that once took weeks could now be completed in days or even hours.

The Challenge

Researchers faced three key hurdles:

  • Efficiently searching over 40 million documents and multi-modal data.

  • Keyword searches that missed crucial connections.

  • Manual cross-referencing prone to errors.

The Solution

Deep Lake Research UI across radiological data and PubMed articles.

The solution needed to handle diverse data, deliver quick responses, and support a growing research team. Deep Lake by Activeloop enabled AI-powered, accurate search across multi-modal data. Intel’s hardware ensured high speeds for data ingestion, embedding quantization, and querying.

Researchers could now use a conversational AI assistant that analyzed queries, connected diverse data to LLMs, and delivered precise answers with citations across relevant articles and even correlate research findings to patient data.

Cross-referencing data across multiple datatypes and clouds, instantly

Solution Benefits

  • Faster Insights: Projects that previously took months now took days, accelerating the research process.

  • Improved Accuracy: Researchers could quickly find connections across clinical trials and device blueprints, uncovering insights that were previously missed, with an average 7% improvement in search accuracy.

  • Enhanced Efficiency: The AI assistant streamlined workflows, reducing manual cross-referencing and minimizing errors, while delivering fast AI searches and enabling seamless connection of any data to LLMs for enhanced analysis.

Multi-Layered Solution for Scientific Discovery with GenAI

Results Achieved with Intel

The impact included:

  • Search times reduced to 0.0243 seconds, achieved using 5th Gen Intel® Xeon® processors for real-time embedding inference.

  • 65% faster literature ingestion, made possible by Intel® Gaudi® 2 accelerators for improved batch processing.

  • Up to 4x faster streaming computations, utilizing Intel® oneAPI Math Kernel Library (oneMKL) for enhanced cosine similarity computations.

Working on a similar research process at your company? Chat to our GenAI experts today to learn how you can leverage all your data for GenAI-powered search.

Mikayel on behalf of Activeloop

© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.