External-To-Internal Research with Human Review Loop

This use case demonstrates how cortx enables a hybrid research workflow by orchestrating both external public LLMs and internal private LLMs, along with human-in-the-loop reviews for compliance. It’s ideal for scenarios where research must be validated and refined before internal use, especially in sectors requiring high accuracy, such as Healthcare, Finance, or Education.

What This Use Case Covers

• Coordinating external + internal research using agents
• Scraping and compiling insights from public LLMs into structured reports
• Sequencing multiple agents with maker-checker approvals
• Ensuring compliance through human validations

Phases Involved

• Phase 1: Workflow Initialization
  1. User logs in to cortx.
  2. Navigates to the Agent Center via the left-hand navigation bar.
  3. Starts a new conversation and selects the Workflow Builder Agent.
  4. User prompts:
     • “Build a workflow with two nodes—first is a Deep Research Agent to analyze sales of office supplies globally, second is an Internal Research Agent to extract supporting insights from internal databases and create a final report.”
  5. The Workflow Builder Agent:
     1. Interprets the prompt.
     2. Designs a 2-node agent workflow with:
        1. Node 1: Deep Research Agent (external research)
        2. Node 2: Internal Research Agent (private data extraction)
     3. Generates a sequence diagram and shares it with the user.
     4. Seeks user approval via maker-checker format for each node.

• Phase 2: Execution (Human-in-the-Loop Enabled)
  • Node 1: Deep Research Agent
    1. Pulls sales data on office supplies from global public sources via external LLM (e.g., OpenAI).
    2. Summarizes insights.
    3. User acts as checker, reviews the output, and approves it.
  • Node 2: Internal Research Agent
    1. Uses prior summary and queries internal LLMs and private databases.
    2. Creates a consolidated report combining external and internal insights.
    3. User reviews and approves final output.

• Human-in-the-Loop Compliance Layer
  • This workflow enforces a 2-phase maker-checker validation:
    • Maker: Agent performs the task.
    • Checker: Human user reviews and validates before progressing.
    • Ensures compliance, traceability, and data integrity in critical research scenarios.

Agents Involved

Outcome

• Final report includes vetted insights from both external and internal sources.
• All steps reviewed and approved by the user.
• Ensures compliance in regulated domains through controlled automation.

Best Practices

• Use this workflow when merging public knowledge with internal intelligence.
• Always include human review loops for research validation.
• Ideal for teams that need multi-source synthesis with auditability.