IMAGE
Five specialised AI agents working in concert — autonomously ingesting, enriching, categorising, and acting on financial data with 93% RAG accuracy and sub-100ms latency.
Managing personal finances involves dozens of disconnected, manual tasks — downloading statements, tagging transactions, calculating spend, researching investments. This platform automates all of it, end to end, with a coordinated swarm of specialised AI agents.
Built on LangChain and LangGraph, five autonomous agents handle the full financial intelligence pipeline: ingesting live bank data via Plaid, enriching and categorising every transaction with LLM reasoning, generating personalised investment recommendations through a RAG pipeline, and executing approved trades via Alpaca — all with hallucination safeguards at every boundary.
The result is a platform that delivers 2.5% monthly ROI with 99.9% reliability, 93% retrieval accuracy, and sub-100ms response times — turning raw bank data into actionable financial intelligence with zero manual intervention.
Connects to Plaid API to fetch live bank transactions, account balances, and financial data. Normalises and routes data downstream to enrichment.
Augments raw transaction records with merchant metadata, category inference, and semantic labels using an LLM chain — turning bank strings into structured intelligence.
Classifies enriched transactions into granular spending categories. Detects anomalies, recurring charges, and subscription drift with zero manual tagging.
Analyses spending patterns and risk profile against market data to generate personalised, context-aware investment recommendations via a LangGraph multi-step reasoning loop.
Routes approved investment recommendations to Alpaca API for automated trade execution — with position sizing, risk guardrails, and real-time confirmation.
Five specialised LangGraph agents collaborate in a directed acyclic workflow — each owning a distinct domain, with clean handoffs and shared state management.
Retrieval-Augmented Generation with Azure Vector Search and Cosmos DB delivers 93% retrieval accuracy — ensuring recommendations are grounded in your actual financial history.
Multi-layer validation at every agent boundary — confidence scoring, source citation requirements, and fallback chains prevent fabricated financial advice.
Connects to real bank accounts via Plaid Link — syncing live transactions, account balances, and institution data with automatic refresh cycles.
Alpaca API integration with configurable position limits, stop-loss guardrails, and portfolio rebalancing logic — 2.5% monthly ROI with 99.9% execution reliability.
ARM template provisions the entire stack — Azure Functions backend, Static Web App frontend, Cosmos DB, and Vector Search — with managed identity auth and zero secrets in repo.