Finance

AI Financial Assistant for WealthFluent

Akveo built an AI assistant that explains complex financial data in plain language while staying compliant with strict US financial regulations. A separate deterministic calculation engine ensures every financial figure is based on verified computations, eliminating AI hallucinations.

Finance

United States

24 months (Ongoing)

AI Development

SaaS Development

[ Client ]

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.

15+

Specialists Involved

6

Months to AI Launch

100%

Regulatory Compliance

[ Challenge ]

Making Finance Easy to Understand Without Compromising Compliance

The client’s platform already offered powerful portfolio tracking, goal planning, and market analytics. The challenge was making all that financial data easy to understand. That is why Wealthfluent was searching for a solution that could explain complex information in plain language while staying compliant with strict US financial regulations.

Making Financial Data Clear and Compliant

The client wanted an AI assistant that would guide without crossing the line into regulated financial advice. Off-the-shelf AI models weren’t a viable option because they can hallucinate, generate inaccurate figures, or present misleading information.

Balancing Speed and Accuracy

The assistant also needed to respond quickly using live financial data without sacrificing accuracy. Every response had to be grounded in reliable, up-to-date information so users could trust the insights they received.

[ Solution ]

Building the Solution

Within 6 months, we built a secure AI assistant on Amazon Bedrock and integrated it into the WealthFluent platform.

Hybrid Intelligence Architecture and Zero AI Hallucination

To solve the biggest technical challenge, we used a hybrid architecture. The AI is responsible for the conversation while a separate deterministic calculation engine handles all the math. This means the AI never invents numbers.

We designed the assistant around a simple principle: guidance, not advice. It educates, explains concepts, and answers questions in plain language, but it never tells them what financial decisions to make. The final decision always stays with the user, allowing the platform to remain fully compliant with regulatory requirements.

Smart Data Preprocessing for Increased Speed

The assistant needed to work with live transaction data in real time without compromising accuracy.

To achieve this, we added a smart pre-processing layer that filters and summarizes account data before it reaches the AI. This allows the assistant to answer questions about portfolio activity almost instantly, while keeping every response grounded in accurate, up-to-date data.

AI Assistant Key Capabilities

The assistant supports users in three key ways:

Conversational onboarding. Instead of a static tutorial, it guides new users through onboarding with a natural conversation, helping them set up their profile and financial goals while reducing the need for human support.
Proactive insights. It identifies spending patterns and highlights opportunities to adjust budget, surfacing useful insights before users even ask.
Contextual education. Users can highlight any financial term in the chat to receive a clear explanation sourced from a curated knowledge library rather than the internet.

Technical Implementation

At its core, the AI assistant runs on Amazon Bedrock with Anthropic Claude, while LangGraph orchestrates the conversation flow and business logic. The application itself is built with Angular and Node.js.

To meet the client's security and compliance requirements, the solution runs within their own AWS environment.

The delivered solution comprises three core parts:

a modular frontend ecosystem
a TypeScript-based AI assistant for platform interactions
a Python-based portfolio optimization engine for complex financial analysis

1. The Modular Frontend Ecosystem

To integrate AI without disrupting the existing platform, we built the assistant as an independent frontend using Angular, Angular Elements, PrimeNG, and RxJS. Packaged as a single JavaScript bundle, it can be embedded and updated independently via an automated CI/CD pipeline.

Two-Way Platform Integration	Performance & Scalability	User Experience & Multimodality	Security & Enterprise Readiness
A custom two-way event-driven architecture enables the AI assistant to consume real-time context and trigger actions directly in the app, powering interactive experiences.	Using WebSockets, we enabled real-time streaming, persistent sessions, and background processing, allowing users to switch conversations without interrupting long-running AI tasks.	The interface supports text and voice interactions, document uploads, rich Markdown rendering, drag-and-drop positioning, and a full-screen mode.	Enterprise-grade authentication, session management, error handling, chat history export, and slash command support ensure a secure and extensible platform.

2. Core AI Assistant Backend

We built the AI backend in TypeScript on AWS to integrate seamlessly with the client's existing platform while delivering the security required for real-time financial applications.

Multi-Agent Architecture	LLM Orchestration	RAG Pipeline	Real-Time Performance
We used a multi-agent architecture where specialized agents handle specific domains (e.g., transaction analysis).	A custom orchestration layer routes AI requests to AWS Lambda services, integrating existing business logic while allowing independent scaling.	Retrieval-Augmented Generation grounds responses in company documentation and proprietary financial methodology, ensuring accurate answers.	Secure WebSockets enable persistent sessions, while data pagination and context summarization allow efficient processing of large financial datasets within LLM token limits.

3. Portfolio Optimization Agent

Akveo developed a dedicated Python-based AI agent that uses LangGraph to optimize portfolios against target benchmarks, generating transparent buy, sell, and reallocation recommendations.

Terraform, AWS Lambda Layers, and Langfuse provide scalable deployment, dependency management, and end-to-end AI workflow monitoring.

Agentic Workflow Orchestration	Intelligent Model Routing	Modular Financial Intelligence & Dynamic Prompting	AI Governance & Guardrails
Using LangGraph and LangChain, we designed a stateful, multi-step workflow that validates data and refines recommendations until it reaches an optimal outcome.	Dynamic routing across Anthropic Claude models on Amazon Bedrock balances performance and cost, while background context compression enables long-running optimization sessions.	A modular architecture activates specialized financial logic only when needed, with dynamic XML-structured prompts improving efficiency, reasoning consistency, and token usage.	Constraint-based guardrails enforce investment rules and portfolio restrictions, ensuring consistent, explainable, and compliant recommendations.