A small fintech company was stuck doing manual research every time clients asked financial questions. Their advisors had to dig through data sources and rely on memory to piece together responses. I built them a retrieval system that pulls together client data, relevant knowledge base content, and structured analysis so their team can deliver faster, more consistent advice with clear supporting evidence.
The Problem
Every client inquiry meant starting from scratch. The support team would manually pull client data from different systems, hunt through financial documents, and lean on whoever happened to know the most about that particular situation. The whole process was slow and inconsistent — the quality of advice depended too heavily on which person was available and what they remembered from previous cases. Leadership was rightfully nervous about bringing AI into financial advisory work, where a hallucinated recommendation could cause real damage to client relationships and regulatory standing.
Solution & Approach
I built a multi-layered RAG system that separates data retrieval from analysis to reduce hallucination risk. When the team inputs a client issue, the system runs two parallel workflows. One path enriches their query in both Hebrew and English to improve retrieval across their multilingual knowledge base. The other path pulls the client's business and financial data and structures it into clear, factual signals. Only after both data gathering phases complete does the advisory agent synthesize the retrieved knowledge with the client's actual situation to generate recommendations. Every response includes the supporting evidence and source citations, so the team can verify the reasoning and trust that suggestions are grounded in real information rather than model speculation.
Outcome & Results
The team now handles client inquiries with systematic support instead of starting from zero each time. They can access relevant knowledge and client data through a single workflow that produces evidence-backed recommendations in minutes rather than hours. The structured approach eliminated the inconsistency problem while giving leadership the transparency they needed to feel confident about AI-assisted financial advice.
Tech Stack
Interested in building something similar?
No commitment. Just a conversation about your systems.