About LepiVerse
LepiVerse is the Lepidoptera-focused interface powered by BioCosmos, an AI engine for exploring biodiversity through images, traits, taxonomy, geography, and natural-language search.
Natural history collections contain an extraordinary record of life on Earth. They preserve specimens, images, labels, traits, geographic records, and taxonomic knowledge accumulated over centuries. Yet much of this information remains difficult to search, compare, and synthesize at scale. BioCosmos is being developed to make biodiversity data more usable by connecting museum specimens, biological images, taxonomic information, traits, and occurrence records through modern machine learning and interactive web tools.
The project began with butterflies as a model system. Butterflies provide a visually rich and scientifically important group for developing tools for image-based search, trait discovery, fine-grained biological retrieval, and interactive exploration of biodiversity. LepiVerse is the interface that brings these capabilities to the Lepidoptera community, with BioCosmos as the engine behind it. From this foundation, BioCosmos aims to support broader discovery across organismal groups and natural history collections.
What BioCosmos does
BioCosmos helps users move through biodiversity data in ways that are closer to how researchers, students, and collection users actually ask questions. Instead of searching only by scientific name or catalog fields, users can explore organisms by visual similarity, color, morphology, traits, geography, and natural-language descriptions.
The platform is being developed to support questions such as:
- Which species look visually similar to this specimen?
- What butterflies are blue and occur in the United States?
- Which specimens match a particular visual pattern, trait, or ecological context?
- How can museum images be organized into useful biological neighborhoods?
- How can AI help researchers discover, curate, and compare biodiversity data?
BioCosmos combines computer vision, vector search, biodiversity informatics, and language-model-based query tools to make natural history data easier to search, interpret, and reuse.
Why it matters
Biodiversity science increasingly depends on the ability to work across large, heterogeneous datasets: specimen images, occurrence records, taxonomic databases, trait descriptions, and collection metadata. These data are powerful, but they are often scattered across systems and difficult to query together.
BioCosmos addresses this gap by building AI infrastructure for organismal biology. Its goal is not to replace expert knowledge, but to amplify it: helping researchers find patterns, generate hypotheses, identify overlooked specimens, and move more fluidly between images, names, traits, and places.
People
Core team
Arthur Porto
Principal Investigator
Florida Museum of Natural History, University of Florida
Project lead for BioCosmos, with a focus on computer vision, biodiversity AI, organismal biology, and museum-scale research infrastructure.
Heru Handika
Lead Developer
Florida Museum of Natural History, University of Florida
Main developer of the BioCosmos platform, including core software infrastructure, model and data integration, search systems, UI/UX design,deployment workflows, and user-facing tools for AI-powered biodiversity discovery.
Jose Fortes
Co-Principal Investigator
University of Florida
Co-PI of the UF AI² award supporting BioCosmos, with expertise in natural language processing, AI systems, and research computing infrastructure.
Venkata Suresh Yarava
DevOps and Research Computing
University of Florida
Supports deployment, infrastructure, system administration, and the computational environment needed to run BioCosmos as a scalable research platform.
Michael Elliot
Large Language Models
University of Florida
Contributes to the language-model components of BioCosmos, including natural-language querying, model integration, and AI-assisted interaction with biodiversity data.
Moritz Lürig
Butterfly Dataset Development
University of Florida / BioCosmos collaborator
Led the assembly, cleaning, and organization of the butterfly image dataset that serves as a foundational resource for developing, evaluating, and demonstrating BioCosmos.
Oliver Dobon
Machine Learning Researcher
BioCosmos / BioVision Lab
Developed the VLM research framework with a focus on encoding biological priors into fine-tuning objectives for fine-grained retrieval.
Kira Nichtawitz
Undergraduate Researcher / Intern
Florida Museum of Natural History, University of Florida
Contributes to the design of the web layout, data integration, and visualization.
Georgia Tech and HAAG collaborators
BioCosmos benefited from early prototyping support from researchers in Georgia Tech's Human-Augmented Analytics Group (HAAG). HAAG researchers were primarily involved in the initial prototyping phase of the project, helping shape early work on data ingestion, frontend and backend infrastructure, vector search, deployment, model evaluation, and LLM-assisted query control.
Breanna “Bree” Shi
Georgia Institute of Technology / Human-Augmented Analytics Group
Instrumental in organizing and guiding the HAAG contribution to BioCosmos.
Thomas Deatherage
Georgia Institute of Technology / Human-Augmented Analytics Group
Contributed to the initial BioCosmos prototyping effort.
Romouald Dombrovski
Georgia Institute of Technology / Human-Augmented Analytics Group
Contributed to the initial BioCosmos prototyping effort.
Elan Grossman
Georgia Institute of Technology / Human-Augmented Analytics Group
Contributed to the initial BioCosmos prototyping effort.
BioCosmos is an active research and software-development effort, with contributions from students, developers, and researchers working on model evaluation, user interfaces, data pipelines, image embeddings, trait discovery, and biological search.
Funding and support
University of Florida AI² Seed Award
BioCosmos: A Foundational Multimodal AI Model for Biodiversity
2024–2026 · PI: Arthur Porto · Co-PI: Jose Fortes
University of Florida Biodiversity Institute
Support for biodiversity AI, natural history collections research, and the development of computational infrastructure for studying biological diversity at scale.
University of Florida Research Computing
Hosting and research-computing support for the BioCosmos platform.
Data and responsible use
BioCosmos is built to support research, education, and biodiversity discovery. The platform integrates public biodiversity data, museum specimen information, and biological images where available. Because AI models can reflect biases in training data, sampling, taxonomy, geography, and image availability, BioCosmos should be used as a tool for exploration and hypothesis generation rather than as a substitute for expert verification.
We welcome feedback from biodiversity researchers, collection managers, students, educators, and developers interested in improving AI tools for natural history collections.
Contact
For questions about BioCosmos, collaborations, or research use, please contact: