Launching EVA: The World’s First Commercial VLM Service on Rebellions NPU​

In close collaboration with Rebellions, the EVA team has continuously advanced the technology stack and successfully built a production-grade NPU-based runtime environment for EVA. We are now moving beyond technical validation and officially entering the phase of commercial service deployment.

1. ATOM-MAX NPU Performance Validation: A New Standard for VLM Inference (As-Is)​

EVA recently evaluated operational feasibility with the Qwen3 VL 8B model on Rebellions' latest ATOM-MAX NPU environment. This was not just a benchmark for model accuracy, but a validation of key operational requirements for real industrial services.

• Rebellions ATOM / Qwen3 VL 8B / Accuracy 0.7996 / F1 0.6733
• GPU A100 / Qwen3 VL 8B FP8 / Accuracy 0.7779 / F1 0.5979

Compared with GPU (A100), EVA achieved equivalent or better performance on overall inference metrics. In particular, in fire and smoke detection scenarios, the NPU environment demonstrated stronger processing capability, proving applicability to high-complexity industrial safety monitoring.

2. Optimization and Stability for Commercial Operations (As-Is)​

In real-world deployments, AI systems must handle far more than clean benchmark inputs. Mixed text-image requests and multiple simultaneous camera streams can easily create bottlenecks. To address this, EVA has continuously improved optimization at both the NPU compiler and system levels.

It is critical to build a resource orchestration framework that efficiently distributes CPU, memory, and NPU workloads, so multiple AI Agents can run concurrently without performance degradation. It is equally important to resolve unexpected failures and ensure stable, uninterrupted operation when text-only and image-analysis requests arrive at the same time.

• Complex data processing stabilization: We fully resolved potential malfunctions in multi-core environments where Text Only and Text + Image requests are mixed, significantly improving operational reliability.
• Resource efficiency optimization: By precisely controlling data processing policies across CPU, memory, and NPU, we achieved a high-efficiency runtime where multiple VLM instances can run simultaneously without inference speed degradation.

3. Throughput Optimization Based on Parallel Architecture (To-Be)​

EVA is also pushing forward full-stack parallelization to maximize the multi-core architecture of Rebellions NPU and further advance end-to-end technology integration.

• Parallelization strategy: We are developing techniques to remove VLM inference bottlenecks by applying data parallelism (DP) to the Vision Encoder and tensor parallelism (TP) to the Text Decoder.
• Integrated operations strategy: We are defining the optimal number of concurrent instances and core allocation ratios across multiple NPU resources. This enables GPU-level throughput while significantly improving performance-per-watt and reducing TCO (Total Cost of Ownership).

Closing: The Commercial Era of Efficient Industrial AI​

The combination of EVA and Rebellions NPU is not a simple hardware replacement. It represents a full-stack transformation toward always-on AI inference in the field with predictable operations and a strong balance of high performance, high efficiency, and high stability. Based on validated NPU optimization technologies, EVA will accelerate digital transformation in industrial environments with a more cost-efficient operating model.

Related Reading​

• EVA on NPU Technology Roadmap
• EVA X Rebellions Technology Collaboration

The government’s “2026 Economic Growth Strategy” is reshaping the safety management paradigm across industrial sites​

The core message is clear.
Generous tax incentives will be provided for the adoption of safety facilities leveraging new technologies such as AI,
while severe penalties—at a level that could threaten corporate survival—will be imposed for violations of safety obligations.

This policy is not a recommendation.
By presenting both clear incentives for safety investment and devastating penalties in the event of accidents,
the government is making safety investment a mandatory requirement rather than an optional choice for companies.

Three Key Pillars of the Policy​

• Three-Tier Tax Incentive Package
  Expansion of tax credits up to 12–40% for investments in AI-powered safety facilities
  (R&D tax credit rates: General 2–25%, New Growth & Core Technologies 20–40% / Investment tax credit rates: General 1–10%, New Growth & Core Technologies 3–12%)
  

• Stronger Accountability
  Introduction of new penalty fines equivalent to 5% of operating profit or 3% of revenue in the event of fatal accidents (Legislation and amendments planned for the first half of 2026)

• Intensified Oversight
  Expansion of industrial safety inspectors to 2,095 personnel and introduction of restricted bidding for high-risk construction projects

---

Safety Investment Is Now State-Supported​

The government has begun recognizing AI-based safety facilities as “New Growth & Core Technologies.”

This signals more than regulatory relaxation—it reflects a shift in perception,
where safety technology investment is now regarded as a core pillar of national competitiveness.

• AI Monitoring Systems Included in Integrated Investment Tax Credits

Tax credit eligibility is significantly expanded to include AI-powered intelligent CCTV monitoring systems,
even beyond legally mandated facilities.
The EVA solution represents the standard for next-generation safety facilities recommended by the government.

• Preferential Treatment for R&D and Facility Investments Under New Growth & Core Technology Designation

Investments in advanced safety technologies utilizing AI and robotics are eligible for tax credits
ranging from up to 12% for facilities to 40% for R&D.
(R&D tax credit rates: General 2–25%, New Growth & Core Technologies 20–40% / Investment tax credit rates: General 1–10%, New Growth & Core Technologies 3–12%)
Now is the optimal time to deploy high-performance EVA solutions at the lowest possible cost.

---

A Single Accident Can Undermine an Entire Company​

Regulatory scrutiny has become significantly more stringent.
The introduction of a punitive penalty framework starting in 2026 makes reactive, after-the-fact responses no longer viable.

• Penalty fines of 5% of operating profit or 3% of revenue in the event of fatal accidents (cap: KRW 100 billion)

Under amendments to the Occupational Safety and Health Act and the Construction Safety Special Act,
companies facing repeated fatal accidents may be fined up to 5% of operating profit or 3% of revenue (legislation and amendments planned for the first half of 2026).

This goes beyond a simple fine—it poses a direct threat to corporate management and sustainability.

EVA intelligently monitors worksites 24/7, proactively identifying and blocking fleeting risk factors that human operators are likely to miss.
Accident prevention is now the most reliable cost-saving strategy.

---

Why EVA: A Full-Stack AI Safety Solution​

More than just an AI solution, EVA delivers a full-stack AI experience optimized for each customer’s business environment.
Enjoy unparalleled operational efficiency and flexibility—free from dependence on any single technology stack.

1. Freedom to Choose AI Models Optimized for Your Environment​

Select the AI stack best suited to your operational environment while maintaining powerful safety monitoring performance.

Global leading AI models such as HyperCLOVA X, Qwen, Gemma, LLaMA, ChatGPT, and EXAONE,
along with cutting-edge vision models including ultralytics YOLO, OWL-ViT, RT-DETR, and OmDet,
can all be freely deployed on the EVA platform.

2. One-Quarter of the Total Cost of Ownership (TCO) Compared to Competitors​

High deployment costs and slow maintenance hinder growth.
EVA delivers overwhelming cost efficiency—from initial investment through ongoing operations.

Comparison Criteria	mellerikat EVA	Conventional AI CCTV Solutions
Initial server requirements (100 CCTV cameras)	1 high-performance GPU server (NVIDIA L40S), initial cost approx. KRW 25M	4 high-performance GPU servers (NVIDIA L40S), initial cost approx. KRW 100M
Additional cost for scenario changes	No additional license required, no ongoing maintenance cost	Additional cost for every scenario change, approx. KRW 200K per update
Scenario configuration & modification	Directly editable by operators, under 30 seconds	Requires specialized personnel, approx. 1–2 weeks

Now Is the Right Time. Let EVA Be Your Trusted Safety Partner.​