Secure AI Powered Financial Assistant for WealthFluent

WealthFluent (formerly Ripsaw) is a U.S.-based fintech company democratizing financial planning. Their platform blends portfolio tracking, goal setting, and self-directed investing — and they partnered with us to transform it into an intelligent financial companion.

‍

About the Client

WealthFluent (formerly Ripsaw) is a US-based fintech company. They are on a mission to democratize financial planning. By combining portfolio tracking, goal setting, and self-directed investment tools, they help users control their financial future. WealthFluent needed a partner to transform their platform from a passive tracking tool into an intelligent companion.

Service Type: AI Development, SaaS Development

4+ Specialists Involved
6 Months to AI Launch
100% Regulatory Compliance

‍

Challenge

The client approached us with a clear vision: Make finance feel friendly.

Their platform was powerful, but users felt overwhelmed by charts and numbers. WealthFluent wanted to bridge this gap with an AI companion that could explain complex data in plain English.

However, they faced two critical hurdles:

Trust & compliance: The AI needed to provide helpful insights without crossing the line into "financial advice," which triggers strict legal regulations.
Data overload: The system had to process massive amounts of live transaction data in real-time. A slow chatbot would kill the user experience, but a fast one risked hallucinating (making things up).

They needed a partner who could balance speed, security, and absolute reliability.

‍

Strategy

We moved beyond standard chatbot integration to build a true Financial Intelligence Engine.

Our pillars:

Guidance over advice: We engineered the AI to act as an educator. It analyzes data and offers educational insights, empowering users to make their own decisions while keeping the platform compliant.
Hybrid intelligence: We combined the creativity of large language models with deterministic logic. This means the AI writes the friendly text, but a calculator does the math — ensuring numbers are always 100% accurate.

Seamless onboarding: Instead of a boring tutorial, the AI interviews new users, helping them set up their profiles and goals through a natural conversation.

Building the Solution

We designed a modular, multi-agent architecture built around two specialized backends and an independent frontend ecosystem.

‍

Frontend (Angular + WebSockets)

The chat widget was built as an autonomous micro-frontend using Angular Elements. They are compiled into a single JS file and embedded into the host platform as a plug-and-play module.

A Custom Events system enables two-way communication: the agent pulls live context from the parent app and pushes actions back to update dashboard widgets in real time. WebSocket-powered token streaming keeps sessions alive in the background, supporting long-running optimization workflows and multi-thread chat sessions.

‍

Core AI Backend (TypeScript + AWS)

The primary assistant runs on AWS using a multi-agent collaborator pattern. It is a specialized sub-agents handle distinct domains (transactions, goals, financial planning) and report to the orchestrator.

A custom Chat Tool Router decouples tool logic from the agent, standardizing how it interacts with dozens of AWS Lambda functions. Context is managed through dynamic pagination and summarization to handle large MySQL datasets within token limits. RAG grounds responses in the platform's technical documentation and a proprietary knowledge base.

‍

Portfolio Optimization Agent (Python + LangGraph)

A dedicated Python agent handles investment optimization via cyclic LangGraph workflows. AI Agent revisits and refines results across multiple iterations until portfolio goals are met. Intelligent model routing sends complex reasoning tasks to Claude Sonnet while delegating lightweight queries and continuous context compression to Claude Haiku, balancing capability with cost.

An Agent Skills module dynamically loads specialized prompt modules (e.g., Tax Location Optimization) on demand, preventing context bloat. Guardrails enforce constraint-based rules, which include restricted holdings, diversification targets, liquidity floors. The rules keeping the AI within professional boundaries. The entire environment is provisioned via Terraform, with a self-hosted Langfuse instance providing full traceability of every decision step.

User empowerment

Users now receive proactive data-driven hints. The system identifies spending patterns and suggests budget adjustments automatically.

‍

Operational efficiency

The AI handles 100% of initial onboarding, guiding users through complex setup processes that previously required human support or static FAQs.

‍

Innovation velocity

By using Amazon Bedrock, we established a flexible foundation. This allows WealthFluent to swap in newer, faster AI models as soon as they are released without rewriting code.

‍

How We Built a Secure AI Financial Assistant for WealthFluent

Client overview

About the Client

Challenge

Strategy

Seamless onboarding: Instead of a boring tutorial, the AI interviews new users, helping them set up their profiles and goals through a natural conversation.

Technology stack

Solution