NVIDIA and iSoftStone are making it simpler and faster to develop, deploy, and operate AI
At iSoftStone, we’re proud to be an NVIDIA partner with expertise across its software and hardware offerings. From delivering complex AI enablement projects to building a new generation of gaming laptops to developing AI infrastructure for enterprise, our 20+ global delivery centers work daily to bring the power and possibility of NVIDIA’s technology to clients.
Here in the US, our team can tap into that deep NVIDIA expertise and share the best skills and innovations with you.
Innovations like MAREK, the AI infrastructure for enterprise solution that makes it simpler and faster for companies to develop, deploy, and operate AI.
Built on NVIDIA AI Enterprise, MAREK was jointly created by iSoftStone’s APAC team and NVIDIA’s technical team for one of the world’s largest multi-level marketing companies. Here’s how that client is using MAREK to overcome technical challenges, realize its AI transformation, and respond nimbly to market challenges.
iSoftStone brings the AI expertise and specialized talent enterprises need to move fast — and smart.
Jason Kui, Vice President, Retail Services | iSoftStone
AI transformation is hard. AI infrastructure for enterprise is the answer
The client was feeling it, and we’ll say it: AI transformation is a headache.
It’s the downside of disruption. Technology iteration is rapid (it’s called the AI revolution for a reason!), AI models are always advancing, and tools can be quickly superseded.
If that wasn’t enough, the technology stack of AI application development is fundamentally different from traditional software development. It requires new processes, more automation, new engineering capabilities, and careful resource management.
At odds with this is day-to-day business reality: Maintaining your competitive edge. Nobody wants to max out their R&D budget only to be locked into a stale services partnership or a redundant platform.
True AI transformation requires the right foundation for innovating safely and iterating quickly. It’s about investing in strategic resourcing, not just collections of AI tools. For many enterprises, that starts GPUs and infrastructure.
Making that strategic shift can feel monumental.
Our client was grappling with all these concerns. Its technology team wanted robust AI infrastructure so they could be nimble and dynamic with evolving AI technologies. They wanted to support scalable, secure AI across business functions, develop their own AI applications and tools, keep their data in-house, and better manage costs.
In short: the team wanted a true strategic shift from “general computing with AI tools bolted on” to AI-first computing at the heart of the enterprise.
In the MAREK AI infrastructure for enterprise solution, they found the right way to make it happen.
MAREK, an AI infrastructure for enterprise solution co-created with NVIDIA
To enable the shift to intelligent computing, NVIDIA and iSoftStone brought together AI models, development, evaluation, deployment, operation, and maintenance in one full-stack solution.
- NVIDIA DeepSeek R1 NIM allows the client to self-host GPU-accelerated inferencing microservices and access pre-optimized AI models and industry-standard APIs, so its team can build everything from chatbots to agents flexibly.
- NVIDIA NeMo provides access to state-of-the-art AI models and supports model optimization, custom development, and full lifecycle management.
- NVIDIA NeMo Guardrails defines, orchestrates, and enforces guardrails to manage compliance and strengthen security.
- NVIDIA NeMo Curator for Developers supports real-time data collection, processing, evaluation, publishing, model training, and knowledge base construction.
- NVIDIA Base Command Manager Essentials (BCME) automates the full AI lifecycle and streamlines training, deployment, and monitoring.
- NVIDIA Run:ai enables clients to schedule and virtualize GPU resources dynamically across AI workloads—ensuring efficient use of compute for tasks like model training, Copilot deployment, and high-concurrency inference—while reducing operating costs and ensuring scalability.
MAREK is designed for on-premise/on-cloud intelligent computing resource management and allows cloud resources to be integrated later as needs grow and long-term AI development goals shift. The client adopted a “minimum start-up” approach, choosing on-premises to avoid vendor lock-in and validate use cases while retaining full control over their data, systems, and governance, with plans to adopt a hybrid approach in the future.
MAREK’s support of customized data pipelines and real-time processing ensures the AI models can make inferences based on the freshest information. Privacy protection mechanisms in processing, desensitizing, and encrypting sensitive data comply with laws like GDPR as well as the client’s internal security requirements.
The results for this client? Support for the entire AI pipeline with AI computing that’s secure and stable, allows for cost-effective innovation, and empowers their team to scale rapidly whenever they want to.
Rapid AI rollout at scale, measurable ROI
With the right infrastructure in place, the client had everything it needed to support its AI strategies and empower its business users.
The client’s technology team was soon testing use cases, including Copilots and AI assistants with low latency and personalized outputs that were also capable of handling millions of real-time frontline sales and marketing queries per day.
And deployments were faster, allowing tools to be rolled out rapidly yet safely. With MAREK, the content review for one of the AI marketing assistants took only a couple of days to finalize instead of multiple weeks.
MAREK’s use of NVIDIA DeepSeek R1 Nim also helped the client achieve a 40% efficiency improvement in inference performance, while NVIDIA BCME resulted in a 60% increase in AI ops efficiency.
The speed and efficiency benefits from AI were soon felt at the business level too. For example, an internal-facing AI office assistant that automated document tasks, generated PowerPoints, and helped with data extraction gave the team using it an impressive 50% boost in efficiency.
Need strategic, secure, scalable AI? We’ve got you
MAREK is an AI orchestration framework we designed with NVIDIA Blueprints in mind — flexible, scalable, and ready for deployment across any enterprise.
Looking for something else? From AI strategy and consulting to AI planning to AI deployment, iSoftStone is ready to help. Whether you need a different AI infrastructure solution or a proof of concept for a tool, our services are highly customizable. We’ll tap into the global power of 90,000+ employees to source the right blend of expertise and experience for your unique needs.
