Leverage AI within Oracle Cloud Infrastructure (OCI) for PeopleSoft

Enterprise technology is evolving from static menus to conversational, purpose-driven interfaces. For organizations using Oracle PeopleSoft, the focus has shifted from whether to adopt Artificial Intelligence to how to implement it securely and effectively.

Building on earlier initiatives such as PICASO, it is now essential to address current requirements for scalability and advanced reasoning. This has led to a more robust approach: integrating PeopleSoft with the OCI AI Agent Platform and Generative AI (GenAI) Service.

Transitioning to a natural language assistant requires three core components working together:

1. PeopleSoft Environment:

The user interface enables users to submit questions, while backend services such as Integration Gateway and Search Server store and manage the data.

2. OCI AI Agent Platform:

The orchestration layer uses Large Language Models (LLMs) to interpret user goals and map them to specific tools.

3. OCI GenAI Service:

This engine delivers the advanced reasoning needed to convert human language into technical queries.

Unlike traditional chatbots with rigid, predefined dialog flows, Agentic AI operates autonomously. These agents decompose complex objectives, plan actions, and interact with external APIs to deliver comprehensive solutions. Leveraging the OCI AI Agent Platform, PeopleSoft customers gain enterprise-grade security, deep business data integration, and scalable, mission-critical deployments.

A high-performance AI assistant must distinguish between search and action tasks. The recommended architecture maintains a clear division of responsibilities:

1. Read-Only Operations (PeopleSoft Search)

For queries such as "Who are the remote workers reporting to me?" the assistant uses PeopleSoft Search, powered by OpenSearch.

• Performance: Accesses flattened data indexes, avoiding impact on online transaction processing (OLTP).
• Security: Enforces existing user access controls at the search level.
• Mechanism: An "NL to Search Query Service" converts the user's natural language into a secure OpenSearch API query.

2. Transactional Operations (REST APIs)

For actions such as approving expenses or updating records, the assistant uses standard OpenAPI-based REST services through the PeopleSoft Application Services Framework. These services manage business logic for inserts, updates, and deletes, ensuring data integrity.

As AI architectures expand, maintainability becomes increasingly challenging. The Model Context Protocol (MCP) offers a unified layer that standardizes communication between AI agents and business tools.

• Portability: Enables tools to interoperate across different platforms.
• Composability: Makes it easier to chain complex workflows together.
• Governance: Provides auditable interactions, meeting strict enterprise compliance requirements.

The NL to Search Query Service is the core engine enabling users to retrieve structured business data through conversational language. It serves as a bridge, converting a user's natural language question into a secure, high-performance OpenSearch query.

Below is a deep dive into the end-to-end implementation process:

1. Intent Mapping and Category Retrieval

The process begins by identifying the functional area relevant to the user's question.

• Data Structure: The service fetches a list of available search categories and descriptions (e.g., Vouchers, Requisitions, or Remote Workers).
• LLM Analysis: The user's question and the category list are sent to the OCI GenAI Service. The LLM identifies the intent, or functional category, and maps it to the specific search index required to answer the question.
• Intent Accuracy: Descriptions provided to the LLM must be precise to ensure the model accurately identifies the user's goal.

2. Prompt Construction using Sample Queries

To ensure the LLM generates valid queries, the service applies prompt engineering.

• Generic Prompts: These include global instructions, such as formatting dates consistently or returning only the OpenSearch query without conversational filler.
• Index-Specific Prompts: These are tailored to each search index and include sample queries and unique constraints for that data type.

3. Generating the OpenSearch Query

Once the intent is identified, the service creates a well-structured API request.

• Field Mapping: The service retrieves index metadata, including field names, types, and aliases, directly from OpenSearch using the _mapping API.
• Formatting Rules: The LLM is instructed to use the match_phrase construct, keep field names uppercase, and avoid escaping characters or adding line breaks.
• Example Output: { "query": { "bool": { "filter": [ { "match_phrase": { "Display Name": "abcd" } } ]}}}

4. Secure Execution and Result Return

The finalized query is executed against the PeopleSoft Search Server.

• Security Context: All searches honor PeopleSoft's robust security and access controls. The request is sent with the current user's context (e.g., via SearchUser in the HTTP header) to ensure they only see data they are authorized to access.
• Data Handling: The results are returned as a JSON response.
• Formatting: The system can use the LLM to convert this JSON into an HTML table for the user or apply a client-side converter within the chat interface.

5. Continuous Improvement

The architecture includes a feedback mechanism, typically a thumbs up or down icon. This feedback is stored to refine the prompt store, enabling the system to generate more accurate responses over time based on real-world usage.

By adopting an agentic architecture on OCI, PeopleSoft customers can move beyond basic chatbots to develop intelligent assistants. This approach enhances the user experience while maintaining the robust security and performance standards required by enterprise organizations.

Ready to explore how AI can transform your PeopleSoft environment? Connect with our team to start building a smarter, more efficient enterprise experience.