Say Hai: Security & Compliance Overview

A transparent overview of the security, privacy, and compliance measures implemented for the AI-powered ESS Assistant within the Employee Self Scheduling (ESS) platform.

1. General Information

The ESS Assistant enhances the Employee Self Scheduling platform by providing intelligent support and guidance to employees through conversational AI.

The integration uses Microsoft's Semantic Kernel framework to securely connect backend services with a Large Language Model (LLM). The currently selected model is OpenAI GPT-4.1, chosen following internal evaluation of performance, reliability, and security.

AI technologies are continuously evaluated and improved. Any future model upgrades will undergo structured assessments to ensure security, compliance, and operational reliability.

2. High-Level Architecture

The ESS Assistant operates as a secure layer between employees and the ESS platform.

Backend communication takes place through controlled APIs, ensuring that all permissions, authorizations, and security boundaries remain identical to those of the standard ESS application.

3. User Interaction Flow

Employees interact with the assistant using natural language.

The assistant interprets user requests and securely retrieves the required information through authenticated backend integrations and predefined plugins.

4. Model Access & Permissions

Token-Based Authentication

• All backend calls require a valid access token.
• The LLM does not possess, store, or manage access tokens.
• Permission validation is identical to standard ESS user interface interactions.

Endpoint Restrictions

• Only approved ESS Assistant API endpoints can be used.
• The LLM cannot directly call OWS APIs.
• Functionality is provided exclusively through controlled plugins.
• Available plugins only expose functionality permitted for the authenticated user.

No Direct Database Access

• Data remains within the customer's environment.
• Data retrieval takes place exclusively through authenticated APIs.
• The LLM cannot query databases directly.

5. Conversation Context & Caching

To provide natural and coherent conversations, Redis is used to temporarily maintain conversation context.

• Prompts and responses are stored in Redis.
• Conversation history is supplied only for the current user.
• Cached context expires one hour after the last interaction.
• Users cannot access chat histories belonging to other users.

6. Logging

User interactions are logged to support platform monitoring, troubleshooting, and continuous improvement.

• Logs are written in a write-only manner.
• The LLM has no read access to logs.
• Logs are retained for 90 days.
• After 90 days, logs are automatically and permanently removed.

7. Microsoft Data Processing Commitment

Microsoft contractually commits that customer data submitted to Azure OpenAI is not used to train foundation models.

Additional information regarding Azure OpenAI data processing, storage, and privacy controls can be found in Microsoft's official documentation:

Azure OpenAI Data Privacy Documentation

Key Security Principles

• Role-based access control remains fully enforced.
• No direct database access for the LLM.
• Token security matches standard ESS authentication.
• User conversations are isolated per user.
• Temporary context retention with automatic expiration.
• Write-only logging with controlled retention.
• Enterprise-grade privacy protections through Azure OpenAI.