About the Agent

ZBrain's Job Description Update Agent automates the validation and revision of enterprise job descriptions using a Large Language Model (LLM). By integrating directly with Oracle Fusion HCM or a similar enterprise system and aligning job content with internal rule sets from a connected knowledge base, the agent ensures each job description is accurate, compliant, and ready for publishing, without the need for manual intervention. It intelligently updates only the non-compliant sections, preserving the original tone, structure, and role intent, while generating transparent summaries of applied changes.

Challenges the ZBrain Job Description Update Agent Addresses

As job roles evolve and hiring criteria shift, enterprises struggle to keep job descriptions up to date across departments. HR teams often rely on manually reviewing and editing JDs stored in systems like Oracle Fusion, which is time-consuming, inconsistent, and error-prone. Many JDs miss required skills, outdated terminology remains unchecked, and compliance guidelines are often overlooked. Existing workflows rely heavily on subject matter experts or hiring managers for validation, creating bottlenecks in the recruitment cycle. Traditional tools lack the contextual awareness to assess whether a JD meets internal standards and regulatory criteria without overwriting important content.

ZBrain Job Description Update Agent eliminates these challenges by integrating directly with Oracle Fusion HCM or a similar system to extract current job descriptions and validating them against role-specific rules sourced from a connected enterprise knowledge base. It uses an LLM to identify non-compliant or missing elements, revise only the necessary sections, and generate a complete, updated job description. The agent also provides a structured compliance checklist and a summary of applied fixes, ensuring transparency and auditability. With standardized, policy-aligned outputs ready for review and system integration, the agent helps HR teams reduce manual workloads, accelerate recruitment readiness, and scale job description governance with confidence.

How the Agent Works

ZBrain Job Description Update Agent streamlines the end-to-end process of validating and updating job descriptions by integrating Oracle Fusion data, enterprise rule sets, and LLM-powered logic. Below, we break down each step, from raw input through to final delivery, and highlight the key tasks and outcomes at every stage.

Step 1: Input Submission and Classification

The process begins when a user submits a request containing a Job Title, Opportunity Number, or Opportunity ID via an interface or webhook.

Key Tasks:

• Trigger Activation: The agent is activated through a webhook whenever an input is received.
• LLM Classification: A LLM interprets the input and classifies it into one of three categories—Job Title, Opportunity Number, or Opportunity ID.
• Routing: Based on the classification, the agent routes the request to the appropriate Oracle data retrieval path.

Outcome:

• The agent ensures the correct processing path is initiated, allowing for accurate data retrieval from Oracle Fusion regardless of input format.

Step 2: Oracle Job Data Retrieval

After the input is classified by the LLM, the agent determines the type of input provided—Opportunity ID, Opportunity Number, or Job Title, and follows a tailored Oracle API sequence to retrieve complete job data.

• Case 1: If the input is an Opportunity ID (Direct Retrieval):
  • Triggered when: The user provides a valid Oracle Opportunity ID directly.
  • Key Tasks:
    • API Call: The agent makes a direct GET request to the Oracle HCM Recruiting Opportunity API using the provided Opportunity ID.
    • Job Data Retrieval: Oracle returns a detailed JSON object containing all relevant job posting fields.
    • Custom Code Processing: A JavaScript function parses the job object and extracts structured fields such as:
      • job_title
      • description
      • responsibilities
      • qualifications
      • recruiter
      • hiring_manager
      • location
      • organization
      • requisition_number
      • publish_date
  • Outcome:
    • Complete and accurate job data retrieved instantly without additional processing.
• Case 2: If the input is an Opportunity Number (Two-step Resolution):
  • Triggered when: The user provides a job number (e.g., “50037”) but not the internal Opportunity ID.
  • Key Tasks:
    • Initial Lookup: The agent performs a POST to Oracle’s indexed job search endpoint using the Opportunity Number as the query.
    • Opportunity ID Extraction: The search results include the internal Opportunity ID corresponding to the job number.
    • Secondary API Call: A follow-up GET request is sent using this derived Opportunity ID to fetch full job details.
    • Data Normalization: As in case 1, the custom code block processes the response and extracts structured job data.
  • Outcome:
    • The Opportunity Number is first used to look up the corresponding Opportunity ID, followed by the retrieval of full job details in two steps.
• Case 3: If the input is a Job Title (Fuzzy Search & Semantic Matching):
  • Triggered when: The user submits a Job Title (e.g., “HR Analyst” or “Lead Cloud Architect”).
  • Key Tasks:
    • Indexed Search Request: A POST request is sent to Oracle’s indexed search API using the input title as the keyword.
    • Result Set Filtering: The search may return multiple results due to similarity-based scoring (e.g., a search for “Analyst” may return “Data Analyst,” “Business Analyst,” and “Analyst Supervisor”).
    • LLM Ranking: A dedicated prompt-based LLM reviews the title list and selects the best semantic match based on:
      • Exactness of match
      • Functional relevance
      • Contextual intent (e.g., "HR" vs. "IT")
    • Opportunity ID Extraction: The best-matched result’s Opportunity ID is extracted.
    • Final API Call: The ID is used to perform a GET request to fetch the full job description and related metadata.
    • Job Data Structuring: The response is parsed and normalized via the custom code block.
  • Outcome:
    • Enables flexible, user-friendly input while ensuring precise job role identification through intelligent semantic analysis.
• Case 4: If the input cannot be classified or resolved:
  • When triggered: None of the input types (ID, Number, or Title) could be confidently classified or resolved.
  • Key Tasks:
    • Validation Check: The input is re-evaluated for partial matches or recognizable patterns.
    • Exit:
      • If the agent cannot classify or resolve the input, it halts downstream execution.
      • Returns an appropriate error message.
  • Outcome:
    • Ensures that unclear or invalid inputs are caught early, reducing false processing and enabling the user to correct their request.
• Final Outcome:
  • Regardless of the input type, the agent produces a complete, structured, and validated job data object. This object is passed to the next stage for knowledge base rule matching and compliance evaluation.

Step 3: Knowledge Base Rule Retrieval

After the job data is extracted from Oracle, the agent initiates a title-based search against the enterprise knowledge base to retrieve role-specific validation rules. The way this title is obtained depends on the input type used in the earlier step.

• If the input was an Opportunity ID:
  • Key Tasks:
    • The job title is extracted directly from the Oracle job detail response using the Opportunity ID.
    • The agent queries the enterprise knowledge base with this job title.
    • The KB returns role-specific rule sets, which may include:
      • Required skills
      • Years of experience
      • Certifications or educational background
      • Organizational or regional compliance criteria
  • Outcome:
    • An accurate role-specific rule set is retrieved, with no ambiguity, as the title is directly mapped from Oracle’s official job record.
• If the input was an Opportunity Number:
  • Key Tasks:
    • The agent performs a preliminary Oracle search using the Opportunity Number to retrieve its corresponding Opportunity ID.
    • Then it performs the same Oracle job detail lookup as in case 1 using that ID.
    • The title from the Oracle response is used to query the KB.
    • A rule set specific to that title is returned.
  • Outcome:
    • This path mirrors case 1 after resolution. A clean job title from Oracle leads to accurate rule retrieval with no manual disambiguation needed.
• If the input was a Job Title:
  • Key Tasks:
    • The agent submits a fuzzy title match request to Oracle's index search endpoint.
    • Oracle returns a list of close matches ranked by semantic similarity (not exact match).
    • An LLM processes the list and selects the best match, returning:
      • Selected Title
      • Matched Opportunity ID
    • Using the Opportunity ID, the full job details are fetched from Oracle.
    • The selected title is then used to query the KB for corresponding rules.
    • If the title is ambiguous or generic, the agent uses an LLM to:
      • Filter irrelevant rule sets
      • Consolidate similar ones
      • Resolve inconsistencies
  • Outcome:
    • Despite a more uncertain input, the LLM ensures the agent retrieves the most semantically aligned title and an appropriate rule set for that role.

Step 4: LLM-based Validation and Revision

After retrieving the job data and applicable rules, the agent invokes an LLM to validate the description and revise it if needed.

Key Tasks:

• The agent packages two inputs for the LLM:
  • The structured job data from Oracle
  • The validation rule set from the Knowledge Base
• The LLM evaluates whether each field (e.g., qualifications, responsibilities, tools) satisfies the rules.
• Validation Logic:
  • If a field is compliant, it is retained without modification.
  • If a field violates a rule, the LLM revises only that field, preserving original tone, intent, and structure.

Outcome:

• A validated, policy-aligned version of the job description is generated with contextual edits applied only where needed.

Step 5: Final Processing and Delivery

Once the job description has been validated and revised by the LLM, the agent prepares and delivers the final outputs to the user.

Key Tasks:

• Formatted Output Generation:
  • The LLM returns a formatted Markdown output containing:
    • Job Validation Checklist: A tabular breakdown of each rule with compliance and fix status.
    • Summary of Fixes: Clear, human-readable explanation of what was changed and why.
    • Updated Job Record (JSON Format): A complete, restructured job description in machine-readable format.
• Output Storage:
  • The full Markdown output is stored in the agent’s runtime state using a key-value storage block. This allows it to be retrieved later in the final output stage.
• Webhook/API Response:
  • The final response is returned to:
    • The calling webhook (if triggered via API)
    • The integrated application (e.g., HR dashboard, documentation system)
    • The agent dashboard

Outcome:

• The user receives a finalized, compliance-checked job description in both human- and machine-readable formats. This can be directly published, further reviewed, or piped into downstream systems with zero manual cleanup required.

Why use Job Description Update Agent?

• Rule-aligned Content: Ensures job descriptions consistently align with organization-defined standards and evolving role requirements.
• Accuracy and Consistency: Delivers accurate, standardized, and role-aligned JDs across departments.
• Reduced Manual Effort: Minimizes HR workload by automating comparison and revision tasks that would otherwise require SME intervention.
• Faster Updates: Accelerates the process of keeping job descriptions current, especially during role changes or hiring surges.
• Audit-Ready Output: Produces a revision checklist to track compliance and provide a transparent view of content changes.
• Improved Hiring Accuracy: Enhances candidate-job fit by maintaining up-to-date, standards-aligned descriptions.
• Scalable Implementation: Can handle bulk updates across numerous roles, making it ideal for large enterprises and fast-scaling organizations.
• System Agnostic Scalability: While currently integrated with Oracle Fusion, the agent can be adapted to other job data systems with minimal change.