By Dave DeFusco
At the recent 2025 IEEE/ACM CHASE Conference, Lakshmi Priya Ramisetty, a 2024 graduate of the M.S. in Artificial Intelligence, introduced a new kind of artificial intelligence鈥攐ne not built for tech giants or billion-parameter showdowns, but for veterinarians. Her presentation, 鈥淰etMamba: Optimizing Veterinary Language Modeling with Mamba Architecture for Long-Sequence Efficiency,鈥 revealed a powerful, domain-specific AI model designed to process the long and complex medical texts common in animal healthcare.
Most language models today are optimized for human medical records, product reviews or internet chatter, but veterinary medicine despite its scientific rigor remains an underserved field in the AI world. Ramisetty, who works in AI, ML and Graph technologies at BlueArc, saw an opportunity to close this gap by designing an AI system that understands the language of veterinary care.
鈥淰eterinarians face a flood of information and a shortage of tools to handle it,鈥 she said. 鈥淰etMamba is our answer to that problem.鈥
Natural Language Processing, or NLP, is the branch of AI that focuses on teaching machines to read and write like humans. It powers everything from chatbots to digital assistants. But while NLP has revolutionized healthcare applications for people by summarizing patient records, suggesting diagnoses and even answering medical questions, it hasn鈥檛 made similar progress in veterinary medicine.
Part of the issue is data. Veterinary texts are long, unstructured and filled with specialized terminology. Another problem is the technology itself. Most large language models use a standard AI system called a 鈥渢ransformer,鈥 which becomes slower and uses more memory as the text gets longer, making it hard to handle lengthy veterinary documents efficiently. They also struggle with the kinds of multipage records and research articles common in veterinary settings.
Instead of using transformers, Ramisetty built VetMamba on a newer design called the Mamba architecture. This approach uses a more efficient way of processing information, called a state-space model. The key advantage is that it reads and analyzes text in a more straightforward, step-by-step manner. As the text gets longer, the system鈥檚 workload grows steadily鈥攔ather than ballooning out of control, as it often does with older models. That makes VetMamba especially good at handling the long, detailed records common in veterinary care鈥攓uickly, accurately and without requiring massive computing power.
Asked how often to vaccinate a cat, VetMamba responded: 鈥淭he American Kennel Club (AKC) recommends vaccinating your cat every six months.鈥 In another query about nutrition for a growing puppy, it advised: 鈥淚 would recommend a low-carb diet. This is because the puppy is growing at a very rapid rate, and the body needs to be able to use the energy it is getting from the food.鈥
鈥淭hese aren鈥檛 rote responses pulled from a database,鈥 said Ramisetty. 鈥淭hey reflect the kind of reasoning and contextual understanding that could make VetMamba a valuable assistant for veterinary professionals, vet tech students or even pet owners looking for trustworthy advice.鈥
Veterinarians often work in understaffed environments, with limited time to read through dense case histories or keep up with the latest literature. VetMamba could help by summarizing long documents, answering complex clinical questions or assisting in the education and certification process.
鈥淰eterinary medicine deserves the same technological support as human medicine,鈥 said Ramisetty. 鈥淲e built VetMamba to make that possible.鈥
