Unveiling PlantScience.ai: A niche LLM to power your plant science research
You ask a plant science specific question to a general-purpose large language model (LLM) and it puts out vague answers. Sometimes you wonder if the answer is valid or if the LLM is pulling you into a hallucination trap! Does this sound familiar? To tackle this problem, PlantScience.ai, a virtual scientist for plant sciences, has been rolled out recently as a domain-specific LLM for plant sciences. It provides the much-needed source credibility by providing citations from which the information was retrieved. Built using an automated scientific knowledge graph (AutoSKG) construction approach, this AI assistant adopts a dynamic learning approach. It continuously integrates latest research findings into the existing knowledge database by programmatically querying open access Application Programming Interfaces (APIs) and processes only CC-BY licensed publications. The performance of the AI assistant was assessed by domain specific human experts as well as another LLM on the grounds of factual accuracy, relevance and comprehensiveness, and it proved to perform relatively well compared to the general-purpose LLMs like Gemini, and Opus. Further, with the rapid developments in the AI era, the line between ease of information access and data privacy is becoming negligible. PlantScience.ai uses a client-side storage of all chat history to uphold the data privacy and security aspects, allowing its use in one’s own unpublished or sensitive research projects. After passing a yearlong set of tests by plant scientists at the John Innes Centre, and the Sainsbury Laboratory, PlantScience.ai is now available for use at https://plantscience.ai. Moreover, the AutoSKG pipeline can be used to construct knowledge graphs from one’s own unpublished work or protected data and the source code for it can be accessed at https://github.com/COLA-Laboratory/autoSKG Thus, this work can serve as a reference point to create such domain-specific LLMs for other niche fields grappling with a large amount of information overflow. (Summary by Shakunthala Natarajan @shakunthalan.bsky.social) Molecular Plant. 10.1016/j.molp.2026.03.010
