Wealth AI: Institutional Portfolio Orchestrator

Wealth AI is an institutional grade, AI-driven wealth management platform. The goal is to provide Relationship Managers (RM) and Financial Advisor with real time, deeply analytical portfolio reviews.

Comment

INSPIRATION

The Singapore Hackathon 2026 challenged us to address a critical gap in modern wealth management: financial advisors struggle to deliver personalized, data-driven investment advice at scale. Legacy tools are fragmented, analysis is manual, and outputs are often generic.

We envisioned Wealth AI—a platform that simulates an institutional investment committee within software. Instead of relying on spreadsheets and intuition, advisors can generate fiduciary-grade insights instantly through orchestrated AI agents and quantitative models.

SOLUTION OVERVIEW

Wealth AI (Institutional Portfolio Orchestrator) is a unified platform that transforms raw client and market data into actionable investment intelligence.

With a single click, advisors can generate comprehensive wealth reports that incorporate multiple analytical perspectives and rigorous quantitative modeling.

Core Capabilities

Multi-agent portfolio analysis across value, safety, and growth strategies

10,000-path Monte Carlo simulations for outcome forecasting

Advanced risk metrics including Sharpe, Sortino, CVaR (95/99), and Max Drawdown

Goal-based planning aligned with client objectives

Smart SHA-256 caching reducing API usage by approximately 90%

Institutional-grade reports highlighting consensus and conflicts

SYSTEM ARCHITECTURE

The platform follows a decoupled, microservices-inspired architecture.

Backend (Python / FastAPI)

LangGraph-based DAG orchestrating 10+ AI and quantitative nodes

Unified LLM access with automatic fallback

Real-time market data ingestion

Permanent Salesforce integration via JWT OAuth

Vectorized NumPy computations for high-performance simulations

Content-addressable caching for cost efficiency

Frontend (React + Vite)

Dark-themed, glassmorphism-inspired dashboard

Client grid with search and risk-based filtering

Interactive visualizations for portfolio and risk metrics

Robust markdown rendering with error boundaries

KEY CHALLENGES AND SOLUTIONS

API Rate Limits: Solved using smart caching, persona batching, and retry control

Salesforce Authentication: Implemented JWT bearer flow for permanent access

Monte Carlo Performance: Reduced runtime from 12 seconds to under 2 seconds using NumPy vectorization

Unstable LLM Output: Mitigated with frontend error boundaries and fallback rendering

Cross-Origin Issues: Resolved via explicit CORS configuration

ACCOMPLISHMENTS

True multi-persona investment analysis at scale

Institutional-level quantitative rigor

90% reduction in API costs through intelligent caching

Reliable, token-less Salesforce integration

Production-grade UI with resilience engineering

Built by a four-member team within 48 hours

KEY LEARNINGS

LangGraph is well-suited for complex, stateful AI workflows

Caching is critical for scalable AI systems

Vectorized computation significantly outperforms iterative logic

Frontend error handling is essential when working with LLMs

Clear team ownership accelerates execution

FUTURE ROADMAP

Fine-tuned, lower-cost language models

Multi-currency portfolio support

Real-time updates using WebSockets

Mobile advisor application

Automated regulatory compliance checks

Additional investment personas and historical backtesting

Built With

Share this project:

Updates