01
Challenge
A production AI financial assistant for Non-Resident Indians navigating NRE/NRO banking, FEMA rules, tax-treaty questions, mutual funds and long-term planning — built as a full-stack platform with a streaming chatbot, RAG knowledge pipeline, personalisation, task tracking, billing and admin. The main delivery challenge was to give users complex financial data explained through a secure and approachable product experience, while keeping Streaming AI chatbot, RAG knowledge pipeline, and Personalisation layer consistent across the complete journey. We framed the engagement around the smallest production-worthy flow first, then expanded the scope without weakening the underlying component and data model.
02
How we built it
We started by separating financial content, user context, and conversational state so each layer could evolve independently. The first production slice connected Supabase authentication and RLS-protected profiles to a streaming GPT-4o chat. We then added the pgvector retrieval pipeline, tuned chunking and retrieval against representative NRI questions, and introduced personalisation only after grounded answers were stable. The final stage connected recommendations to persistent tasks, billing, and an operator panel, with the longer chat workflow on Node.js and fast knowledge retrieval on the Edge runtime.
03
Delivery
The delivered platform is more than a chat interface: it combines grounded financial answers, individual context, and follow-through in one workflow. Users can move from a cross-border finance question to a saved, actionable plan, while the operator team can update knowledge and manage the product without engineering support. The split runtime and typed Next.js/Supabase foundation also leave room to expand the corpus and product surface safely.
