Toyotron

Our system leveraged NVIDIA Nemotron not only as the core reasoning model but also as the multi-agent orchestrator, coordinating specialized sub-agents for Intent, Research, Finance, and Report Generation. We combined this with a Next.js + TypeScript frontend and a modular backend exposing API-based tool endpoints. For trade-in estimates, we integrated an OpenAI vision model that performed image recognition on uploaded car photos, pulled VIN metadata from the NHTSA API, and fused condition scoring with our recommendation pipeline. The agentic architecture handled conversation state, scheduling workflows, email generation, and dynamic task routing through prompt-layered governance and persistent session context. Even within a 24-hour sprint, we shipped a polished, extensible prototype that demonstrates how agentic AI + multimodal reasoning + modern web frameworks can meaningfully elevate the automotive retail experience.

01 / 05