Skip to main content

Using the Chat and Feedback

EVA is designed around a chat-based interface for each camera, allowing users to handle everything—from alert monitoring to configuration control and feedback—within a single flow. Without navigating through multiple menus, users can understand and control the system through conversation, while also evaluating detection results in real time.

As EVA monitors cameras, alerts generated based on scenarios are displayed in real time as messages in the chat interface. Each alert message can be clicked to view in detail, providing an intuitive understanding of the analysis results along with image snapshots.

Additionally, by enabling the “Show Detection Alerts Only” option at the top of the chat window, general conversation messages are hidden, and only detection alerts are displayed. This allows operators to quickly grasp situations without unnecessary distractions.

The chat interface also provides a “Detection Alert History” feature, allowing users to search past alerts. You can filter by specific time ranges or keywords, making it effective for event analysis and issue tracking.



Natural Language-Based Control​

The chat interface is not just for monitoring alerts—it also functions as a control interface for EVA. When users input requests in natural language, the EVA Agent’s LLM (Large Language Model) interprets them and performs actions such as updating settings or providing guidance.

Without navigating complex menus, users can directly access and configure what they need. The system not only explains features but also guides users on how to apply them, enabling faster and more intuitive operation.

For example:

  • "Set detection targets to people and dogs."
  • "Set vehicle sensitivity to 0.85."
  • "Change the AI inference interval to 10 seconds."
  • "Adjust brightness to 10."

Through the chat interface, users can directly control key features such as detection targets, sensitivity, inference intervals, and video settings.

💡 Even if you don’t remember a feature or cannot find a setting, simply ask in natural language to receive immediate guidance.



Scope of Support​

Control via the chat interface supports most EVA features, including both configuration changes and feature guidance.

However, detection scenario generation must be performed through the dedicated Generate process and cannot be created directly within the chat interface.



Providing Feedback​

EVA allows users to provide direct feedback on detection results, and this process is also centered around the chat interface.

Based on alert messages, users can immediately assess the accuracy of results and provide additional explanations through the detailed view if needed.

  • Correct / False Detection Selection Users can quickly indicate whether an alert is correct or incorrect directly from each message.

  • Detailed Feedback Clicking a message opens a detailed view, where users can review images and analysis results while providing more specific feedback.

This structure enables users not only to review results but also to convey the reasoning behind their judgment to the system.



Utilizing Feedback Data​

User-provided feedback is stored along with image snapshots, detection scenarios, correctness labels, and user explanations. This data is managed internally in a Multi-modal Vector Store and is later used to evaluate similar situations.

Rather than serving as simple records, this feedback continuously refines detection criteria for each camera environment. As a result, EVA becomes increasingly optimized for real-world conditions.



Performance Improvement and Analysis​

Accumulated feedback data is used to improve system performance and analyze operational status.

  • Detection Performance Optimization Decision criteria are continuously refined based on correct and false detection data.

  • Quantitative Analysis Reports Labeled detection data enables analysis of metrics such as accuracy and false alarm rates.

This allows operators to objectively assess system performance and identify areas for further improvement.



💡 During the initial camera setup, false alarms may occur due to limited environmental data. With continuous feedback, EVA quickly adapts and becomes optimized for the specific environment.