Case Study • Healthcare AI
A production-grade bilingual patient support chatbot for Spartina that automates routine medical inquiries and hands complex cases to people.
Patient support lines spend most of their day on questions that already have good answers. Those answers existed, but in a curated English knowledge base, while patients ask in Persian.
A chatbot here can't just be friendly. A wrong answer to a medical question is worse than no answer, so the system had to know when to stop talking and call a human.
In medical support, knowing when not to answer is a feature.
An advanced RAG pipeline that bridges language and retrieval, with a clear route out to people.
Keyword and vector retrieval combined, so exact drug names and fuzzy descriptions both find their passages.
A second pass orders candidates by actual relevance before anything reaches the model.
Complex cases are routed automatically to the support team instead of being answered badly.
How a patient's question becomes an answer or a handoff.
Persian queries answered accurately from English sources, without translating the whole knowledge base.
Support response times improved, and manual call-center workload dropped significantly.
Automated escalation means the bot takes the routine work, not the risky decisions.
