To maintain technological autonomy, MISA deployed an on-premise cluster featuring 8 NVIDIA DGX™ H200 systems connected with NVIDIA Quantum InfiniBand networking. This infrastructure powers both training and inference, serving as the engine for MISA’s AI factory. It enables the company to train and fine-tune models ranging from 8B to 235B parameters, while also hosting its “Core AI” directly on the DGX cluster. By bringing Core AI in-house, MISA transitioned from an as-a-service model to a fully self-hosted architecture, running its largest and most intelligent LLMs on DGX systems. Complementing this setup, NVIDIA A40 GPUs handle smaller, specialized workloads such as eKYC and router LLMs.

By utilizing NVIDIA Run:ai, MISA maximized its hardware investment through GPU virtualization and fractional sharing, enabling 10–20 concurrent experiments per week—up from just one or two previously.

MISA leverages the full NVIDIA AI Enterprise software stack to streamline its end-to-end workflow. It uses NVIDIA NeMo™ for domain-specific fine-tuning, ensuring its models excel in Vietnamese and local administrative contexts. For production-scale serving, MISA utilizes NVIDIA Dynamo and NVIDIA NIM™. By leveraging FP8 quantization, inference throughput has increased from 1,200 to 4,500 TPS per model, which is essential for real-time AI agent interactions.

The system features more than 50 pre-built AI agents deployed to serve customers, while enabling enterprises to design and customize their own AI agents according to internal business needs.

Among the pre-built agents are MISA AVA, an AI assistant for internal information retrieval; MISA AMIS OneAI, a unified AI platform for knowledge research and expert responses; MISA AMIS Support agent for 24/7 assistance; MISA AVA Mail for email management; and MISA AVA Recruitment for automating hiring processes.

In addition, MISA provides more than 500 customizable AI agent templates tailored to specific domains, such as sales agents for retail sectors, scheduling agents for service businesses, recruitment agents for HR tasks, and B2B consulting agents. These agents can be adapted to align with unique workflows and industry needs, helping organizations efficiently scale their AI adoption.

MISA’s AI solutions are used by more than 450,000 organizations, demonstrating the scalability, stability, and high availability of the AI infrastructure in large-scale, real-world operational environments.

By deploying an end-to-end NVIDIA-accelerated full-stack architecture, including a cluster of NVIDIA DGX H200 systems (64 NVIDIA H200 GPUs) with NVIDIA AI Enterprise and NVIDIA Run:ai, MISA reduced fine-tuning times by 4x and achieved $200,000 in monthly cost savings. This architecture supports more than 50 AI agents, processing 2.5 million daily feature usages to automate public services and enterprise workflows nationwide.

MISA has realized its goal of embedding generative AI and AI agents into core business workflows through the MISA AVA AI Assistant integrated within the MISA AMIS all-in-one business management platform.

The results demonstrate a clear and measurable impact:

• Finance and Accounting: Previously, businesses spent up to 80% of their time on manual tasks and required up to six accountants to process tens of thousands of invoices monthly. With MISA AVA, tasks such as invoice validation, reconciliation, and data entry are fully automated. Financial documents and reports can now be generated in just a few minutes instead of hours, while eliminating human error.
• Sales and Business Operations: MISA AVA provides 24/7 support, instantly delivering product and pricing information, automatically summarizing calls, assisting with email drafting, and extracting customer data. This allows sales teams to eliminate most manual tasks and focus on customer engagement and revenue generation.
• Recruitment and HR: Previously, each hiring cycle required up to five HR staff to manually process hundreds or thousands of applications. With AI, the entire workflow—from CV intake and screening to evaluation and data extraction—is 100% automated within minutes, saving up to 90% of HR processing time.
• Management and Executive Decision-Making: MISA AVA consolidates financial, sales, and HR data into real-time reports, analytics, and forecasts. Business leaders can now retrieve insights using natural language commands, receiving instant comparisons, trend analyses, and predictive insights that previously required manual reporting.

These outcomes demonstrate that MISA is not only embedding AI for assistance but also advancing toward AI agents capable of autonomously executing and optimizing end-to-end business operations.

The impact of MISA’s NVIDIA-powered AI is visible across Vietnam:

• Enterprise Automation: Small businesses such as Noomfood use MISA AI agents to handle 24/7 customer inquiries and order synchronization. This has increased customer response speeds sixfold, allowing staff to focus on higher-value tasks rather than manual data entry.
• Nationwide Scale: More than 450,000 organizations are enabled with MISA AI capabilities, processing over 1.1 million LLM requests daily.

MISA plans to evolve its AI factory into a comprehensive ecosystem of AI agents that serve both internal operations and external customers. By mastering the full LLM technology stack—from pretraining to model optimization—it is securing its technological autonomy and long-term scalability. Looking beyond text, the company’s road map includes expanding into image, voice, and 3D technologies, providing a high-performance foundation to deliver secure, AI on a national scale.