GSK deploys Undermind to power AI research agents grounded in the scientific record

GSK

Undermind has the greatest impact compared across all GenAI solutions in our scientific space, and is the one all my teams shout the most about.

R&D leader, GSK

Undermind enables GSK to rapidly integrate findings from the scientific literature with our existing data, a key part of our AI strategy.

Kim Branson, SVP and Global Head of AI/ML, GSK

Key Highlights

  • Over 1000 GSK scientists use Undermind in their daily work
  • Undermind integrates with GSK’s internal AI ecosystem as the research-evidence layer of its “AI Scientist” stack.
  • GSK’s surveys and data show enthusiastic and enduring use among their scientists

AI at GSK

GSK is driving an enterprise-wide AI transformation, with capabilities spanning target discovery through clinical development. They are building a full-stack AI approach—where data, models, agents, and evidence systems work together to accelerate science and support better-informed decisions. This stack includes capabilities built in-house at GSK, complemented by collaborations with strategic partners.

Partnering with Undermind

As part of this strategy, GSK partnered with Undermind to bring literature-grounded AI research agents, and AI-powered alert systems about new research, into scientists’ workflows.

What Undermind does for GSK scientists

Scientists face an exponentially growing literature base that makes comprehensive searches impractical. Finding the information scientists need, frequently on niche and complex topics, takes days or weeks of effort and often requires consulting with an external expert in a narrow field.

To address this, GSK scientists use Undermind like an AI expert colleague or librarian assistant – they outline a problem statement or frontier topic, then the AI carefully assesses vast quantities of scientific literature and reports back its findings, shrinking days of effort into just a few minutes.

Undermind’s core algorithms, going far beyond simple retrieval-augmented-generation (RAG) by intelligently following citation trails and searching until exhaustion like a human scientist would, ensure that users receive a complete view of what’s known and a coherent understanding of how it all fits together. Because the AI surveys a vast quantity of evidence at once, it can deliver insights and a synthesis of information beyond the practical reach of a human researcher to assemble by hand.

The results generated by Undermind show GSK scientists exactly which works led to key facts and evidence, grounding quotes to specific papers and showing the citation and temporal evolution of evidence.

Additionally, Undermind continues to monitor the web for new research and proactively alerts users to anything new that comes out related to their topics of interest.

Key Undermind Features for GSK Scientists

Generate insights

Access to an AI expert that has read the relevant literature. Generate custom tables to support your research.

Verify responses

Trace any statement by following in-line citations back to the source paper

Gauge relevance

Quickly grasp if a paper is relevant to your unique question. Sort and filter however you like.

Stay up to date

Get notified whenever relevant papers are published.

Programmatic access to Undermind — part of GSK’s AI Scientist effort

GSK’s internal AI platform also connects programmatically to Undermind via a custom API—incorporating a research-evidence layer into their other tools and information flows. Undermind serves as a pillar in GSK’s “AI Scientist” stack, used by both AI agents and human researchers as a key part of longer and more complex workflows.

GSK AI Architecture Diagram: Undermind literature expert exchanges questions and answers with the AI Scientist Platform, which orchestrates internal data integration and provides research alerts

How GSK scientists are using Undermind

Teams report accelerated timelines, higher quality solutions to problems they face, and outputs shared across teams that align decisions and directly impact presentations and portfolio decisions. A few example use cases include:

Early research

  • Probing hypotheses to rank novel target opportunities and assess emerging mechanisms for therapeutic intervention
  • Rapidly conducting target and indication assessments and surfacing shareable evidence chains to support prioritization

Clinical development

  • Summarizing biology and context of how phenotyping biomarkers have been used, supporting identification of candidates for clinical biomarker assessments
  • Providing information on potential preclinical models and assessing outcome measures (e.g. specific imaging metrics) with linked precedent and citations

Specialized research support

  • Supporting toxicology reviews and safety profiling with focused literature analysis
  • Enabling rapid knowledge acquisition in niche or unfamiliar scientific areas, allowing scientists to quickly understand new domains outside their expertise

Competitive analysis

  • Supporting strategic assessments by rapidly collating evidence that would take weeks to compile manually
  • Providing competitive context and market intelligence to support portfolio decisions and strategic positioning across multiple indications

What GSK scientists are saying

GSK performed an internal survey, confirming enthusiastic and sustained use across the organization, from early-career researchers up through leadership.

GSK’s survey reported a net promoter score (NPS) of 63, well above published benchmarks for enterprise end-user software (where scores for similar tools measured in a workplace setting typically range from around 0 up to the mid-20s), placing Undermind in the top few percent of comparable software.

Results from GSK’s independent survey of over 130 GSK scientists

How likely are you to recommend Undermind to a colleague?

NPS survey results showing strong positive skew toward scores 9 and 10

How would you rate the ease of use?

Ease of use survey results showing 93% rated it easy

How did Undermind meet your expectations?

Expectations survey results showing 88% said it exceeded expectations

Individual experiences shared by GSK scientists

Undermind is absolutely transforming the way we gather scientific information. It is my new way to gather information about any scientific question going forward.

I found it easy to use, and the quality of output on scientific queries was excellent.

Undermind has quickly become a must have.

Instead of wasting days in finding the same (or worse) answers, I can decide faster how to move forward in my project and in particular how to find something new that was never explored.

Undermind’s platform has provided a huge value. Work which would take weeks with an agency takes a few hours with this tool.

Undermind is absolutely transforming the way we gather scientific information. It is my new way to gather information about any scientific question going forward.

I found it easy to use, and the quality of output on scientific queries was excellent.

Undermind has quickly become a must have.

Instead of wasting days on finding the same (or worse) answers, I can decide faster how to move forward in my project and in particular how to find something new that was never explored.

Undermind's platform has provided a huge value. Work which would take weeks with an agency takes a few hours with this tool.

Looking ahead

By eliminating time-consuming literature review and ensuring access to the best insights, Undermind allows GSK scientists to focus on the most interesting work: solving complex scientific problems and pursuing the ambitious research that leads to new therapeutic breakthroughs.

The partnership between GSK and Undermind reflects the importance of building more autonomous research capabilities for modern R&D and how GSK is planning for the future of research.

Undermind enables GSK to rapidly integrate findings from the scientific literature with our existing data, a key part of our AI strategy. By helping our scientists move quickly from question to actionable insight, it accelerates the pace of our discovery of new medicines.

As the volume of scientific data continues to grow, sophisticated approaches to navigate information at scale will become increasingly critical. As GSK continues building their AI capabilities to address this, Undermind serves as an important layer, positioning our research teams to explore novel solutions and identify the most promising research directions.

Kim Branson

Kim Branson

SVP and Global Head of AI/ML, GSK

Interested in learning more?

Connect with our team to explore bringing AI into your research ecosystem.