AutoAdvisor — a multi-agent system for vehicle ownership decisions

The problem

Vehicle ownership cost is deliberately opaque. Lease, finance and CPO offers are quoted on different bases, and the true difference only appears once tax, dealer fees, insurance and maintenance are projected across several years. No single model can hold all three domains well, and buyers were making six-figure lifetime decisions on a monthly-payment number.

What we built

1. Architected a multi-agent system with dedicated specialist agents for New, Certified Pre-Owned and Lease vehicles, so each ownership type is reasoned about by an agent that understands its economics.
2. Designed a central Orchestrator Agent that interprets the query, routes it to the relevant sub-agents, and consolidates their responses into a single comparable answer.
3. Integrated a Vehicle Search Agent over an enriched inventory dataset, so recommendations reference vehicles that are genuinely available.
4. Engineered cost-breakdown models covering the hidden variables — sales tax, dealer fees, insurance and maintenance — projected across a multi-year ownership window.
5. Built the execution layer on AWS Bedrock Agents and Lambda behind a WebSocket API for real-time, streaming interaction.
6. Added AWS Sonic to generate a personalised voice summary of every response, so the recommendation can be listened to rather than read.