3 posts tagged with "AI Agent"

To effectively implement LLM services, a robust LLMOps framework is essential. Among its components, observability (o11y) has evolved beyond simple monitoring to become a critical enabler of the system’s meta-intelligence.

---

The Evolution of o11y into Meta-Intelligence​

Early LLM o11y focused on collecting metrics such as token usage, response time, response content, and user feedback to monitor performance. We adopted Langsmith, a commercial tool, to monitor the execution process of AI logic. Later, we integrated Langfuse, an open-source tool, allowing our organization to selectively use either tool based on licensing requirements.

However, as the number of AI Agent service users grew, it became clear that accumulated data could no longer provide meaningful insights through simple log analysis. Consequently, we decided to transform o11y data from mere "observation logs" into a meta-intelligence tool. This system leverages AI Agent outputs and user feedback to automatically reformulate questions or enhance response quality by adjusting model behavior.

In essence, o11y data transcends real-time performance monitoring to become the cornerstone of a feedback loop that enables AI Agents to self-improve.

Academically, this approach aligns with the growing focus on AgentOps or Agentic AI observation systems. There is a movement to propose comprehensive observation frameworks for AgentOps, tracking various artifacts such as execution paths, internal logic, tool calls, and planning stages. Beyond black-box evaluations, the importance of inferring and optimizing behavioral patterns based on agent execution logs is increasingly emphasized.

---

Mellerikat’s ‘Meta-Intelligence o11y’ Model Architecture​

Our o11y meta-intelligence strategy is structured as follows:

1. Trace-Based Execution Logging Using Langsmith and Langfuse, we capture LLM calls, internal planning, token usage, responses, and feedback at the trace session level.

2. Real-Time Anomaly Detection and Feedback Loop When issues like latency or hallucinations are detected, alerts and automated reports are generated instantly, with an automatic tagging system triggered based on user feedback.

3. Meta-Layer Inference Engine Based on collected data, the system provides intelligent suggestions, such as "Which questions are more appropriate?" or "How should responses be restructured for better effectiveness?"

4. Automated Prompt Tuning and Deployment The meta-intelligence engine automatically refines prompts, and updated logic is applied in real-time through a CI/CD pipeline.

This architecture moves beyond manual tuning of LLM logic, aiming for meta-intelligence where the system continuously proposes and implements improvements.

Today, LLM-based AI Agents are evolving from mere response generators into agentic AI, capable of autonomous reasoning, planning, and execution. In this context, o11y transcends monitoring to become the core infrastructure for self-learning meta-intelligence. Ultimately, o11y is the key to transitioning beyond basic observability into a paradigm where AI Agents continuously evolve, enabling sustained learning Agent services.

For more details on Mellerikat o11y, refer to this link.

EVA Showcases Innovation with Multi-Modal LLM-Based AI Services at Cisco Live 2025​

Cisco Live 2025, June 8-12, 2025, San Diego

Mellerikat participated in Cisco Live 2025, seizing a valuable opportunity to present its innovative AI service, Mellerikat EVA, powered by Multi-Modal Large Language Models (LLMs), to global customers and partners. At this premier event focused on networking, security, and AI technologies, Mellerikat showcased a demo featuring a unique architecture that implements cost-efficient AI solutions, earning enthusiastic responses from attendees. As the first major event following Cisco’s acquisition of Splunk, Cisco Live 2025 highlighted the integrated future of AI and data analytics. Mellerikat unveiled a Multi-Modal LLM-based solution combining Mellerikat EVA, Cisco Meraki Camera, and Splunk Instance through its demo booth, demonstrating practical AI applications in industrial settings. Notably, our innovative architecture, which significantly reduces the operational costs of Multi-Modal LLMs, left a lasting impression on attendees.

---

Integration of Mellerikat EVA and Meraki Camera with Multi-Modal LLMs​

At our demo booth, we showcased an AI service integrating Mellerikat EVA with Cisco Meraki Camera, powered by Multi-Modal LLMs. This solution detected attendees in real time, identified individuals wearing specific clothing, and triggered alerts. Leveraging the strengths of Multi-Modal LLMs, Mellerikat EVA processed both image (camera data) and text data simultaneously, achieving high-accuracy detection. The results were visualized in real time on a Splunk Instance, vividly demonstrating the solution’s scalability and real-time analytics capabilities.

• Core Scenario: Meraki Camera captures footage, EVA analyzes it to detect individuals wearing specific clothing and generates alerts, with results instantly displayed on a Splunk Instance dashboard.
• Cost Efficiency: Multi-Modal LLMs are often considered costly for commercial use, but Mellerikat’s unique architecture significantly reduces operational costs, greatly enhancing industrial applicability.
• Attendee Response: Attendees highly valued the combination of cost efficiency and real-time analytics, recognizing its practical value across industries such as security, retail, and public safety.

This demo exemplified how Mellerikat EVA integrates Cisco’s network infrastructure and Splunk’s data analytics platform to deliver efficient and powerful AI services.

---

Strengthening Technical Partnership with Meraki​

At Cisco Live 2025, discussions with the Cisco Meraki team reinforced the consensus that Mellerikat EVA’s Multi-Modal LLM-based technology is essential for enhancing Meraki’s AI capabilities. The integration of Meraki Camera with EVA was recognized as adding a new dimension of intelligence to network and security solutions through cost-efficient AI services.

• Achievements: Both parties agreed to further integrate Mellerikat EVA into the Meraki platform, solidifying plans for future collaboration.
• Future Outlook: Combining Meraki’s cloud-based network management with Mellerikat’s cost-efficient Multi-Modal LLM technology will enable innovative solutions for industries such as smart buildings, retail, and logistics.

These discussions strengthened the partnership between Mellerikat and Cisco Meraki, significantly expanding opportunities for collaboration in the global market.

---

Pyeongtaek Digital Park Case Study: Recognized as a Top CDA Example​

Mellerikat garnered significant attention at Cisco Live 2025 by presenting its Pyeongtaek Digital Park Country Digital Acceleration (CDA) case study. This implementation of Mellerikat EVA in Pyeongtaek enabled real-time monitoring of workers’ safety conditions and immediate decision-making through alert messages, earning recognition as a top CDA case study.

• Case Overview: EVA was deployed at Pyeongtaek Digital Park to monitor worker safety and send instant alert messages, maximizing operational efficiency. The unique architecture significantly reduced the operational costs of Multi-Modal LLMs.
• Attendee Response: Attendees expressed strong interest in EVA’s industrial applicability, particularly in manufacturing, logistics, and smart factories, appreciating its ability to deliver cost savings and high performance simultaneously.

This recognition as a top CDA case study validated Mellerikat EVA’s ability to create tangible value in industrial settings.

---

The Beginning of Relentless Innovation​

Cisco Live 2025 was a significant opportunity to showcase Mellerikat’s Multi-Modal LLM-based AI service, EVA, on a global stage. Our ability to implement commercially challenging Multi-Modal LLM technology in a cost-efficient manner through a unique architecture resonated deeply with attendees, leaving a lasting impression. The integration of Mellerikat EVA, Cisco Meraki Camera, and Splunk Instance opened new possibilities for digital transformation, while discussions with Meraki and Splunk laid the foundation for future innovations. This event is just the beginning. Through close partnerships with Cisco and Splunk, Mellerikat will continue to develop cutting-edge AI services that transcend the boundaries of networking, security, and data analytics. The enthusiasm and feedback from attendees at Cisco Live 2025 have inspired us to deliver even greater technology and value moving forward.

Background​

Meraki’s Cloud-Managed Service already boasts an exceptional infrastructure. If a variety of third-party apps, particularly AI-based services, could seamlessly integrate with this cloud platform, the true potential for enhancing Meraki’s value could be realized.

Currently, Meraki Cloud includes an App Store with some available apps, but it faces clear limitations:

• Integration with Meraki Cloud Services
  • App installation and deployment are restricted. Only select partners can officially register apps, and the installation process is complex or not automated.
  • Third-party apps are not fully integrated with the Meraki Dashboard, leading to fragmented user experiences or dispersed management points.
  • Limitations in APIs and SDKs hinder sufficient integration and scalability with external services.

Upgrading Meraki Cloud to the next level and establishing best practices for a third-party app ecosystem make the integration of Meraki Smart Camera with mellerikat EVA a highly significant case study.

---

Understanding Meraki Smart Camera & Cloud​

Meraki Smart Camera and Cloud form an outstanding service combination. Far from being a simple CCTV, Meraki Camera aims to be a smart video platform organically integrated with network and security infrastructure.

Key Features of Meraki Camera​

• All-in-One Structure Meraki Cameras have built-in storage (SSD or SD card), eliminating the need for a separate Network Video Recorder (NVR), and support Power over Ethernet (PoE) for power and network connectivity.

• Cloud-Based Management All configuration, operation, and maintenance are centrally managed through the Meraki Dashboard (web-based UI), enabling real-time monitoring and configuration changes without on-site visits.

• Intelligent Analytics Features like motion-based event detection, person detection, and area-specific analytics (e.g., heatmaps, dwell time) provide insights beyond basic surveillance for spatial operations.

• End-to-End Security Stored video data is encrypted during transmission and managed under Cisco Meraki’s security framework, with user role-based access control and audit logs.

---

Core Value of Meraki Camera Cloud​

• Centralized Management Cameras installed worldwide can be managed from a single Meraki Cloud, ideal for large-scale, multi-site enterprises or global chains.

• Instant Video Access While videos are stored locally in the camera, they can be streamed quickly via the cloud anytime, anywhere, with fast and intuitive event or time-based searches.

• Smart Search and Analytics Timeline-based motion searches, person detection-based searches, and cross-camera event tracking significantly reduce the time needed to review footage.

• Hardware Maintenance Efficiency Camera firmware updates are applied automatically, and alerts for anomalies or failures enable proactive responses.

---

True Smart Meraki Camera​

Despite this robust infrastructure, it’s time to consider the next step: delivering True Smart value to customers.

• Cameras should recognize and deeply understand a wider range of scenarios.
• They should enable user-customized scenario recognition and automatically execute desired follow-up actions based on recognized information.
• Data movement, storage, and utilization systems should be optimized to enhance cost efficiency.

These processes should be easily managed and controlled using Large Language Models (LLMs).

For example, cameras could automatically detect specific events (e.g., hazardous situations, congestion, VIP visits) and trigger actions like notifications, automatic door opening, or lighting control. AI could also recommend or optimize video data storage and transfer policies. With an LLM-based natural language interface, queries like “Tell me about the customer who stayed the longest in the store today” can be handled effortlessly.

---

mellerikat EVA + Meraki Camera​

To realize the True Smart vision for Meraki Camera, we propose mellerikat EVA, a Vision AI solution offering the following services.

Features and User Service Scenarios of Meraki + EVA​

• Real-Time Advanced AI Analytics EVA analyzes video feeds from Meraki Cameras in real time, providing services like person, vehicle, and object recognition, behavior analysis, and hazard detection.

• Cloud-Based Seamless Deployment AI capabilities can be activated instantly through integration with Meraki Cloud, without additional hardware.

• Customized User Services Users can easily define desired events (e.g., intrusions in specific areas, queue length, customer behavior patterns), and EVA automatically recognizes and provides alerts for them.

• Demo Scenario For instance, when store congestion increases, EVA can automatically notify the manager and trigger synchronized lighting or audio announcements, enabling immediate operational applications.

---

mellerikat EVA​

EVA is a powerful technical platform for AI services, integrated with various mellerikat technology components.

Technical Strengths of the Backend​

• Modular AI Engine Supports flexible combinations of AI modules for object detection, behavior recognition, face/license plate recognition, and more.

• Continuous High-Quality Service Easily deploys and utilizes new models to consistently deliver high-performance services.

• Auto-Scaling and Distributed Processing Built on a cloud-native architecture, it automatically scales resources up or down based on video analysis workloads.

• API/SDK-Based Scalability Offers RESTful APIs, webhooks, and SDKs for seamless integration with external systems (ERP, POS, IoT, etc.).

• No-Code/Low-Code Interface Provides an intuitive UI, allowing non-experts to easily design and manage AI analysis scenarios.

---

The Importance of mellerikat EVA and Meraki Collaboration​

EVA could partner directly with various camera vendors to provide services, operate as a cloud-based SaaS, or be offered through white-labeling with camera hardware vendors. It could also be installed in a vendor’s cloud or on-premises environment.

However, these approaches face challenges in achieving economies of scale during initial service setup and operations.

• Meraki already has a globally scaled cloud infrastructure and a large customer base.
• Combining Meraki Cloud with EVA enables immediate large-scale service launches and expansion without additional infrastructure.
• Leveraging Meraki’s App Store and API ecosystem automates EVA’s deployment, operation, and updates, maximizing customer accessibility.
• This collaboration eliminates cost barriers to adopting AI in cameras and positions Meraki and EVA to lead the global standard for Vision AI services.

---

Beyond AI​

EVA aims to grow beyond providing AI functions into a technical platform that unlocks diverse possibilities through cameras (Vision).

• Users can train cameras to recognize desired scenarios.
• Recognized scenarios can trigger subsequent actions via the Meraki Control Plane (MCP).
• These processes are effectively managed on an integrated technical platform.

Human behavior can be simplified as “recognition → action,” occurring across physical channels on the organically connected platform of the human body. EVA is evolving to technically replicate such an organic platform.

---

EVA’s Vision: Becoming the AWS of the Smart Camera Market​

EVA aims to transform the on-premises-centric CCTV market into a cloud-based service market. It envisions a future where numerous Vision sensors operate smartly on an interconnected platform, enabling sci-fi-inspired recognition-action scenarios to be implemented and integrated as services.