Back to all tools
Project recordAI & developmentOpen sourceUpdated

Virtual Tyre Engineer

An AI concierge that reads tyre telemetry and turns it into a safety and performance verdict.

An intelligent assistant developed for the Google Concierge Agents track. It ingests real-time telemetry — tyre pressure, tread wear, and heat consistency — to deliver Safety & Performance audits, then hands off to autonomous agents that can source a replacement and book it in.

What does Virtual Tyre Engineer do?

Virtual Tyre Engineer is an AI concierge built for Google's Concierge Agents track at a Kaggle vibecoding capstone, aimed at drivers and fleets who want tyre risk explained in plain language rather than a wall of sensor numbers. It ingests real-time telemetry, tyre pressure, tread wear to the millimeter, and heat consistency, and watches for compounding risk signals, such as a tyre crossing 82% wear alongside a temperature spike and a pressure drop on the same corner, flagging the combination as one critical event rather than three isolated readings. The audit is presented on an interactive 3D chassis view with glowing hotspots over the affected wheel. On a critical event, the agent can hand off into an agent-to-agent workflow that checks stock with a storefront agent, negotiates a price, and books a fitting slot, with spending authorized through a cryptographically signed payment protocol (AP2).

Demo

Overview

Virtual Tyre Engineer is the concierge layer built on top of the Titan Concierge Agent project, submitted to Kaggle's AI Agents: Intensive Vibe Coding Capstone Project under the Google Concierge Agents track. Its job is narrow and practical: watch a vehicle's live tyre telemetry — pressure, tread wear, and heat consistency — and turn raw sensor noise into a clear Safety & Performance audit a driver can act on.

The telemetry layer is built around simulated Michelin SmartWear-style tyre sensors that track tread wear down to the millimeter. Rather than surfacing raw numbers, the agent watches for compounding risk signals — for example, a tyre crossing 82% wear alongside a 115°C temperature spike and a 22 PSI pressure drop on the same corner — and flags the combination as a critical event rather than three isolated readings.

That audit is presented through an interactive 3D chassis view: a wireframe vehicle model with glowing hotspots over the affected wheel, built with Three.js and styled under a dark, high-contrast visual theme. When the agent detects a critical wear event, it doesn't stop at an alert — it can hand off into a companion agent-to-agent (A2A) workflow that checks stock with a storefront agent, negotiates a price, and books a fitting slot, with spending authorized through a biometric, cryptographically signed payment protocol (AP2).

A conversational layer sits on top of the telemetry and logistics stack, powered server-side by Gemini with Google Maps Grounding enabled, so the assistant can also fold in real-world context — routing around a known hazard, or pointing to the nearest fitting bay — while it walks the driver through what the tyre data actually means.

Key features

Real-time telemetry ingestion

Consumes live tyre sensor data — pressure, tread wear, and heat — over MCP, continuously tracking tread wear to the millimeter across all four corners.

Compounding-risk safety audits

Flags critical combinations rather than single bad readings — for instance, a tyre crossing 82% wear alongside a 115°C heat spike and a 22 PSI pressure drop is treated as one urgent event, not three separate warnings.

Conversational concierge interface

A Gemini-powered assistant with Google Maps Grounding explains the audit in plain language and can factor in real-world context like route hazards while recommending next steps.

Interactive 3D hotspot visualization

A Three.js wireframe vehicle model highlights the exact wheel behind an alert, giving the driver a visual anchor instead of a wall of numbers.

Agent-to-agent replacement logistics

On a critical event, the agent can act as an A2A client — fetching a storefront agent's stock card, negotiating price, and booking an immediate fitting slot without manual shopping.

Secure, authorized checkout

Replacement purchases are authorized through a biometric confirmation step and executed via cryptographically signed transactions within pre-approved spending limits, using the Agent Payments Protocol (AP2).

Architecture

Agent OrchestrationA host application coordinating three agent protocols side by side: MCP for telemetry ingestion, A2A for storefront negotiation with an external agent, and AP2 for authorizing payment.
AI ModelServer-side Gemini (gemini-3.7-flash) with Google Maps Grounding enabled, used for the conversational layer and real-world routing context.
Telemetry Data ModelSimulated Michelin SmartWear-style sensor feed per tyre — tread wear percentage, temperature, and pressure — with threshold logic that escalates compounding readings into a single flagged event.
Payment SecurityRSA-2048 asymmetric JWS signing with alphabetically canonicalized payloads, plus unique transaction IDs and timestamps to block replay and double-spend attempts.
InterfaceA dark-themed, Three.js-rendered 3D chassis with hotspots that light up over the tyre driving the current alert.
Node.jsGoogle Gemini API (gemini-3.7-flash)Google Maps GroundingThree.jsModel Context Protocol (MCP)Agent2Agent Protocol (A2A)Agent Payments Protocol (AP2)RSA-2048 / JWS signing

Training programme context

Virtual Tyre Engineer was built for Kaggle's AI Agents: Intensive Vibe Coding Capstone Project, the capstone challenge for Google's AI Agents intensive course, under its Google Concierge Agents track. The brief was to take the course's agent-building concepts — tool use, multi-agent coordination, and grounded reasoning — and turn them into a working agent that solves a real problem end to end, not just a chat demo.

The submission pairs this telemetry and audit layer with a companion storefront agent project, demonstrating a full loop: an agent that reads raw sensor data, decides something is wrong, negotiates with another autonomous agent to fix it, and closes the transaction securely. A demo video walking through the flow is linked alongside the source.

Looking for another tool?

Browse all tools