Katana Studios

Powering Real-Time 3D Automotive Configurator with Multi-Cloud GPU Rendering at Scale

Computer Vision Data Engineering

Empowering real-time product visualization in automotive marketing with a resilient, cloud-native rendering platform and elastic UI Challenge Accepted!

Powering Real-Time 3D Automotive Configurator with Multi-Cloud GPU Rendering at Scale

Process & Story

Katana Studios, a creative technology company specializing in high-end CGI content for the automotive industry, set out to transform the way visual assets are produced and delivered. To support automotive brands in their shift toward real-time, personalized marketing, they envisioned a new kind of platform: a browser-based, cloud-native 3D content creation tool capable of rendering and streaming photorealistic visuals on demand.

The tool, COATcreate, had to enable configurability, collaboration, and rendering at scale, all without sacrificing speed or visual fidelity. Making that happen required more than creative ambition. It demanded a robust, multi-layered technical architecture that could handle compute-intensive GPU tasks, dynamic scaling, cost-effectiveness, and seamless user experiences across devices.

We joined the project as a technical partner, supporting Katana’s internal team with deep expertise in cloud architecture, DevOps, backend engineering, and frontend design. Our mission: help deliver a performant, secure, and scalable real-time rendering platform tailored to the unique demands of 3D workflows in automotive marketing.

Problem

Bringing a high-end 3D content creation tool into the browser isn’t just a UI challenge, it’s an orchestration of multiple advanced systems working seamlessly behind the scenes.

To deliver COATcreate as a production-ready platform for real-time automotive marketing, Katana faced a set of demanding technical challenges:

The platform needed to support low-latency, photorealistic rendering of complex 3D scenes using GPUs, requiring a high-throughput compute environment with minimal delay.
With GPU availability and pricing varying across providers, a flexible, multi-cloud strategy was essential to balance cost, scale, and reliability. The system had to support consistent orchestration across multiple cloud providers like Google Cloud Platform, CoreWeave, and Vultr.
Ensuring responsiveness during peak load required cost-effective scaling of GPU-powered instances. That called for precise control over infrastructure provisioning and coordination.
Supporting large 3D asset uploads meant addressing issues of speed, reliability, and failure recovery, especially over standard browser interfaces.
The inclusion of real-time preview and interaction via WebRTC introduced additional demands on system latency, bandwidth, and synchronization, particularly with global users accessing the platform via Cloudflare’s TURN/STUN services.
The original UI, built on a static canvas, lacked the flexibility and scalability needed for a production tool. A more modular and maintainable frontend architecture was required to support real-time user interaction and future growth.

Solution

To bring COATcreate to life, nexocode joined forces with Katana’s internal team to design and implement a modular, resilient architecture that could support high-performance rendering and live interaction, directly in the browser.

Our contributions spanned cloud infrastructure, DevOps, backend systems, and frontend architecture. Together, we helped lay the foundations for a platform capable of powering real-time product visualization at scale, combining AI-powered 3D content pipelines with a smooth, browser-based user experience.

The result of this cooperation, COATcreate SaaS product, is a browser-based, real-time 3D content creation tool designed to revolutionize automotive marketing. Built upon NVIDIA Omniverse and leveraging OpenUSD, it enables users to interact with CAD data, customize scene variants, and produce photorealistic renders, all within an intuitive and accessible interface.

Outcomes

Real-time rendering powered by high-performance NVIDIA GPUs
Orchestration of multi-cloud GPU infrastructure(with Google Cloud Platform, CoreWeave, Vultr) for optimal performance and cost-efficiency
Development of an asynchronous, microservices-based mechanism for fast and reliable upload of large 3D asset files
Implementation of a custom Go-based communication layer with Kubernetes, ensuring predictable and economical scaling of rendering nodes
A complete overhaul of the frontend, from a static canvas to an elastic, component-driven UI architecture, designed for flexibility and future growth

Case Study Schema Powering Real-Time 3D Automotive Configurator with Multi-Cloud GPU Rendering at Scale

Tools

Google Cloud Platform

React

Nvidia

Vultr

Coreweave

Have a similar project in mind?

LET'S TALK!

Challenges

Orchestrating GPUs across multiple cloud providers
The platform required access to high-performance NVIDIA GPUs, yet availability, pricing, and instance types varied across providers. Relying on a single vendor wasn’t an option. We designed a multi-cloud infrastructure across GCP, CoreWeave, and Vultr, allowing dynamic GPU provisioning depending on cost and availability. This added resilience, but also introduced challenges in managing compatibility, monitoring, and scaling logic across different cloud APIs and GPU specs.

Reliable file upload of large 3D assets
3D workflows often involve gigabyte-sized files, textures, scenes, models. Browser uploads are notoriously fragile, especially at scale. To overcome this, we implemented a microservices-based file transfer mechanism with asynchronous communication and retry logic, ensuring fast and failure-resistant uploads even under fluctuating network conditions.

Scalable GPU node management
Each rendering task consumed significant GPU resources, and traffic was unpredictable. To avoid both bottlenecks and over-provisioning, we built a custom service in Go that interfaced with Kubernetes to control GPU instance scaling intelligently. This allowed the platform to grow and shrink its rendering fleet dynamically, aligned with real-time usage patterns.

UI architecture built for growth
The original frontend was built around a canvas-based prototype, flexible for experimentation, but fragile for production. We led the transition to a modular, component-driven architecture using React, enabling a more elastic interface design that could scale with new tools, layout options, and interaction patterns.

Real-time preview with minimal latency
Delivering live-rendered previews in the browser required seamless WebRTC integration and TURN/STUN routing via Cloudflare. Ensuring smooth video streams while preserving low latency and user control meant careful tuning of bandwidth, session management, and backend synchronization.

Ensuring visibility, stability, and security at scale
Operating a high-performance, GPU-heavy platform in production requires more than code, it needs observability, incident resilience, and operational maturity. We integrated Prometheus, Grafana, and GCP Monitoring, with layered security controls for authentication, authorization, and data protection, ensuring that the system remained secure, performant, and maintainable under pressure.

Our other projects:

Boussias Group

Boussias Group Turning Disconnected Data into a Scalable Data Intelligence Hub

How Boussias transformed manual reporting by unifying 3M+ records and 1,000+ tables into a cloud-native data lake. Challenge Accepted!

Data Engineering

Reverse Vending

Building a Real-Time Packaging Recognition System with Edge-Optimized AI

With sub-500ms detection and lightweight packaging, the computer vision model is built for frictionless scaling, ready to support hundreds of machines in the national deposit return network. Challenge Accepted!

Edge Devices Computer Vision Data Engineering

Orkla Home & Personal Care (OHPC)

Optimizing Procurement through Seamless RFQ Management

The new AI-powered procurement system reduced RFQ processing time by up to 85%, cutting the cycle from 4 weeks to just 4 days and enabling faster, more cost-efficient purchasing decisions.

Challenge Accepted!

Data Engineering Generative AI LLMs