E-Patrol Investigation Agent

What the company appears to be:

The E-Patrol Investigation Agent is an AI-assisted investigation tool built on top of an existing security patrol platform (E-Patrol) developed by it-tude. The agent allows supervisors to ask questions about patrol activity in natural language, using a controlled, read-only data layer and AI orchestration to return findings supported by evidence.

What changed:

The project evolved from an existing operational system into a prototype investigation workspace that leverages AI for querying historical patrol data. It was built during a hackathon (OpenAI 2026) and is described as extending a working platform rather than being based on fictional or synthetic datasets.

Single most important open question:

Is the E-Patrol Investigation Agent intended to be deployed in production environments, and if so, what are the mechanisms for scaling its controlled access and data boundaries beyond the current prototype scope?

The description states that the E-Patrol Investigation Agent is an AI-assisted workspace built on top of an existing PHP and MariaDB-based E-Patrol platform. It uses a separate, read-only investigation architecture to allow supervisors to ask questions about patrol activity in Indonesian or English.

Key components include:

• A dedicated read-only investigation datasource
• Authenticated and fail-closed endpoints
• Tenant- and venue-scoped queries
• Deterministic mapping of patrol and image evidence
• Pseudonymized operator identities
• Sanitized patrol notes
• Bounded investigation tools (e.g., scan activity summary, checkpoint ranking, incident timeline)
• An AI orchestration layer that selects appropriate tools and returns evidence-grounded responses

The system is described as integrated into the E-Patrol interface and available in both Indonesian and English.

Evidence:

• The description states this is an extension of an existing E-Patrol platform.
• It uses a controlled, read-only snapshot with 16,885 patrol records and 338 image records.
• Tools are described as bounded and evidence-aware.
• AI responses include tool used, opaque evidence references, and known limitations.

Inference:

It is an investigative layer built on top of legacy operational data to support human decision-making, not a standalone AI assistant or autonomous system.

The author states that the E-Patrol Investigation Agent was inspired by the need to make sense of large volumes of patrol data collected by existing systems. The core claim is that it turns real-world patrol logs into clear findings and recommended actions using AI, without replacing human judgment.

Key positioning elements:

• It is built on an operational platform already in use.
• It supports investigation through natural language queries.
• It emphasizes trustworthiness by showing evidence and limitations.
• It does not make disciplinary or security decisions autonomously.

Evidence:

• The project was submitted to a hackathon, indicating it’s a prototype.
• The description highlights that the system does not hide uncertainty or invent answers.
• It is described as an extension of an existing product, not a new platform.

Inference:

The positioning reflects a shift from data collection to data understanding, with a focus on explainability and human oversight.

The description states that the E-Patrol Investigation Agent is intended for authorized supervisors in security patrol operations. These users are described as needing to analyze large volumes of patrol activity across multiple venues.

The system supports:

• Supervisors working with real-world patrol data
• Users who need to compare activity between venues or periods
• Teams that require evidence-based findings and limitations

Evidence:

• The agent is for "authorized supervisors" in security operations.
• It supports questions like comparing activity between venues or identifying missed checkpoints.
• It is integrated into an existing E-Patrol platform used by real clients.

Inference:

The target customer is likely a security operations team within organizations using the E-Patrol platform, such as facilities management companies or government agencies with large-scale patrol systems.

Not evidenced.

The description does not mention any pricing model, revenue streams, or commercial arrangements. It only describes the technical architecture and use cases of a prototype built during a hackathon.

The system is described as:

• Built on PHP and MariaDB
• Using a separate investigation architecture
• With controlled, read-only access to data
• Integrated with an existing E-Patrol interface
• Supporting bilingual (Indonesian/English) UI
• Using AI orchestration tools like Codex for development

Key technical features include:

• Tenant-scoped queries
• Deterministic evidence mapping
• Pseudonymized identities
• Sanitized free-text notes
• Fail-closed authentication
• Pagination cursors and truncation reporting
• Evidence-aware AI responses

Evidence:

• The system uses a read-only snapshot of 16,885 records and 338 images.
• It includes tools like scan interval analysis, venue comparison, and checkpoint ranking.
• It avoids unsafe assumptions by using deterministic mapping and explicit limits.

Inference:

The technical design prioritizes data integrity, privacy, and explainability over scalability or automation. The use of Codex suggests a strong focus on engineering collaboration during development.

Not evidenced.

There is no mention of revenue, customers, adoption, or usage metrics beyond the prototype phase. The project was built during a hackathon and described as extending an existing operational platform used by two client venues, but no data on actual deployment or performance is provided.

Not evidenced.

The description does not reference competitors or similar tools in the market for AI-assisted security patrol analysis or investigation systems.

• Prototype scope: The system was built as a hackathon prototype and has no evidence of production deployment.
• Limited data access: The read-only nature and controlled access may limit scalability or adoption.
• No commercial model: No pricing, revenue, or customer data is provided, making it unclear if this is a product in development or an experimental tool.
• Dependency on legacy systems: The system builds on an existing platform with historical data inconsistencies, which could pose long-term maintenance challenges.
• AI trustworthiness: While the system claims to show limitations and evidence, there is no independent validation of its AI accuracy or reliability.

1. Is this project intended for production deployment, and if so, what are the plans for scaling access and data boundaries?
2. What is the current status of the E-Patrol platform it extends? Is it actively used by clients?
3. How does the system handle data governance and compliance in real-world environments?
4. Are there any known limitations or edge cases in how the AI interprets patrol data?
5. Has the team considered integrating feedback from supervisors to improve accuracy or usability?
6. What are the plans for expanding beyond the current 2 venues and 42 checkpoints?