We have started working with academic researchers to test Truleaf's AI research engine on biological research questions. As a plant-science nonprofit and R&D organization, we want people who inspect evidence for a living to judge the engine. They can show us where it helps, where it gets in the way, and what needs to change before it belongs in a research workflow.
We are sharing this while the work is still early. We cannot yet name the researchers, university labs, or questions involved, and we have no results or endorsements to announce. We can explain why we are opening the engine to academic use, how we are approaching that work, and what we need to learn.
Researchers should shape research tools
The engine helps people work through published research and turn scattered evidence into structured, inspectable outputs. It keeps claims connected to their sources, so a reader can move from a synthesis back to the underlying literature. We built it around questions in plant science and biological crop protection, where the answer often depends on the species, conditions, methods, and context.
Testing questions we already understand tells us whether the engine behaves as we expect. It tells us far less about whether it fits the way researchers frame a new question, challenge a source, follow an unexpected result, or decide that the evidence is too thin to support a conclusion.
We think researchers should help shape any tool meant to support research. They apply checks that product development cannot reproduce on its own. They ask what evidence was included, where a claim came from, how uncertainty is represented, and whether disagreement in the literature remains visible. They also catch the awkward moments that a polished demonstration can hide.
We want that level of scrutiny. A smooth interface cannot rescue an output that is hard to inspect or easy to misread. Checking the work has to be part of using the work.
Access is part of our mission
Our nonprofit mission is to make science-backed plant information easier to access. That includes growers looking for reliable guidance and researchers trying to make sense of a difficult body of literature.
Academic teams do not all have the same resources, technical support, or freedom to try new research software. An engine may exist and still be out of reach because access is difficult, its output cannot be examined, or its workflow fits only one kind of organization.
For us, responsible access includes getting the engine into a researcher's hands and making its work inspectable. Researchers should be able to tell where an answer came from and identify uncertainty without digging through layers of presentation. They should be able to use the output as a starting point for their own judgment, then challenge it when the evidence points somewhere else.
Working with academic researchers lets us test those conditions in day-to-day research. It also keeps us honest about the distance between a promising capability and a tool that deserves a place in someone's work.
How we are approaching the work
We start with the research question and the team's current process. We want to understand where they spend time, where the literature becomes hard to navigate, what has to remain visible, and what would make an AI-assisted output unsuitable for their work. The engine may be a poor fit for the problem. Learning that still helps us define its proper scope.
When there is a fit, we ask researchers to work with the engine as it exists, with its current strengths and limits. Feedback grounded in use is more useful to us than feedback on a promised future version. We pay attention to where the engine helps someone move through the evidence and where it adds friction, ambiguity, or extra checking.
The source trail is central to that experience. Biological papers can examine different organisms, environments, protocols, or outcomes while appearing to address the same question. A useful synthesis should make those boundaries easier to see. The researcher still has to decide whether two results can be compared and how narrow a conclusion needs to be.
Our engine is centered on published literature. Biological research also draws on laboratory data, field observations, unpublished findings, specialist methods, and expertise developed over years. An AI-assisted synthesis cannot stand in for that knowledge. The researcher sets the question, examines the evidence, and decides what the material can support. We are building around that division of responsibility.
Privacy and scientific integrity
Early research conversations need room to develop without becoming publicity. We will not identify a researcher, university, lab, or project until everyone involved has cleared that information for release. An exploratory collaboration carries no institutional endorsement and establishes no scientific result on its own.
The same boundary applies to outcomes. If there is work worth sharing later, we will do so only when the people involved agree and we can state the scientific context accurately. Until then, we will keep the details private.
That makes this update less specific than many of our building posts. We can say what has started, what remains unknown, and what we do not have permission to disclose. For now, that is the honest account of the work.
What we want to learn next
We now want to learn which biological research tasks fit the engine while keeping sources traceable, claims inspectable, limits clear, and expert judgment in charge.
Researchers using the engine on questions that matter to them may expose gaps in how we organize or present information. They may ask for context that we have overlooked. Some may find that the engine is the wrong tool for their work. That feedback will help us decide where the engine belongs and where it does not.
If you work in biological research and spend too much time navigating fragmented literature, we would like to hear from you. You can reach our team through Truleaf.org Research and Development. Tell us which parts of a literature review take the most time, what your current tools make difficult, and what you would need to inspect before trusting an AI-assisted output.