Siy.ai

Problem

Reducing administrative work around clinical sessions while keeping the product and AI workflows clearly separated.

What I built

Developed React clinician dashboards and Node.js APIs for session notes, intake forms, assessments, and treatment plans. Built a separate FastAPI service called by the backend over HTTP; its RAG pipeline separated chunking and embedding, indexed vectors in Pinecone, and used OpenAI models for embeddings and context-aware generation.

Service architecture

The Node.js backend calls a dedicated FastAPI service over HTTP, isolating the RAG pipeline from the core application APIs. Within that service, chunking and embedding are separate components: embeddings are indexed in Pinecone, and retrieved context is supplied to an OpenAI text-generation model.

Outcome

The platform aimed to reduce clinician documentation workload by an estimated 50%.

Built with