Abstract:
Pet owners often struggle to interpret symptoms, apply basic first aid, or judge when
a veterinary visit is necessary. In provincial settings such as La Union, Philippines, limited
clinic access and geographic distance compound these gaps. This study designed,
developed, and evaluated VetBot: An AI-Driven Veterinary Chatbot for Canine and Feline
Care Guidance, based on a Retrieval-Augmented Generation (RAG) architecture. The
system was implemented with a Flutter mobile interface, a Django backend, PostgreSQL
for transactional data, and ChromaDB for semantic retrieval, with Gemini 3 Flash as the
language model. VetBot performs triage assessment, delivers general health guidance and
safety-oriented first aid instructions, and provides limited home-based care advice for lowrisk
cases. Supporting modules include pet profiling with a digital vet card, a clinic locator
covering La Union, and accessibility features comprising speech-to-text, text-to-speech,
and bilingual English and Tagalog support.
The RAG pipeline was evaluated with the RAGAS framework on 99 clinical test
queries. Faithfulness scored 0.889 and Answer Relevancy 0.906, indicating that responses
drew on retrieved veterinary sources and addressed user queries directly. Context Precision
(0.753) and Context Recall (0.691) reflected moderately high retrieval performance, with
room for improvement in retrieval coverage and configuration. Answer Correctness scored
0.517; this lower score does not mean VetBot gave wrong answers, but rather that the
metric compared responses word-for-word against a single pre-written reference answer,
so any response that conveyed the same information in different wording was scored lower
even if it was factually correct. Nine (9) licensed veterinarians evaluated VetBot's outputs
on a 5-point Likert scale across nine criteria, yielding an overall mean of 4.56. Triage
Report Utility received the highest rating (4.67); non-urgent scenarios received the highest
urgency-level mean (4.65), followed by urgent (4.57) and emergency (4.47), with all three
urgency levels remaining within the Strongly Agree range. Thirty-two (32) pet owners
completed the System Usability Scale (SUS), producing a score of 79.61 (Grade A-,
“Excellent”). The findings indicate that VetBot delivers reliable, clinically consistent pet
health guidance and connects owners to professional veterinary services without replacing
clinical judgment.
