Case study · Healthcare AI
NeivuX
Multi-Modal Healthcare AI Platform
The brief
NeivuX is a production healthcare AI platform built around a grounded, retrieval-augmented core. A Google Drive-backed knowledge base syncs and chunks documents on an automatic refresh cycle, embeds them with Vertex AI at 1,536 dimensions, and serves answers through Firestore vector search with hybrid reciprocal-rank fusion, AI reranking, and citation-grounded responses that abstain rather than guess. All text AI runs on tiered Vertex Gemini models — a fast tier for everyday work and a pro reasoning tier for the hard questions — on a single Google Cloud stack. Around that core sits the operational layer: streaming chat with a model and reasoning picker, persistent memory extraction, projects, web search, tool calling into operational data such as patient details and billing codes, Billing CX drafting, a quality-audit module, user management with RBAC, a staff time clock, usage reporting, and an embeddable chat widget.
The problem
Healthcare teams sit on large document corpora — policies, protocols, billing rules — but a general chatbot bolted on top invents answers, which is exactly what you cannot ship into clinical or billing work.
Our approach
We built the assistant around a grounded retrieval core: a Drive-backed knowledge base that syncs and chunks on a refresh cycle, Vertex embeddings with Firestore vector search, hybrid reciprocal-rank fusion, and an AI rerank — then wrapped it in operational tooling (RBAC, tool-calling into real data, usage reporting, an embeddable widget).
The outcome
Answers are cited to their source and abstain when the corpus does not support them, rather than guessing. The platform runs the full assistant loop — streaming chat, memory, projects, web search, billing-CX drafting — on a production Firebase stack with a documented eval baseline.
What we built
Tech stack
Key metrics
Want one like this?
We can design, staff, and build a platform like this for your operation.
Get started