The AI Pipeline

Every piece of work produced by a LeanCore specialist passes through a four-stage pipeline. This is what separates a managed team from a chatbot.

The Four Stages

    Your Question
         |
         v
+----------------------------------+
|  1. THE MANAGER PLANS            |
|  Reads the question carefully.   |
|  Thinks about what's really      |
|  being asked. Creates a plan     |
|  for which data to retrieve      |
|  and how to approach it.         |
+-----------------+----------------+
                  |
                  v
+----------------------------------+
|  2. THE EMPLOYEE WORKS           |
|  Follows the plan. Reaches into  |
|  your connected systems. Pulls   |
|  actual data. Assembles an       |
|  evidence-backed answer.         |
+-----------------+----------------+
                  |
                  v
+----------------------------------+
|  3. THE MANAGER REVIEWS          |
|  Checks the work for quality     |
|  and completeness. If it's not   |
|  good enough, sends it back      |
|  for another pass.               |
+-----------------+----------------+
                  |
                  v
+----------------------------------+
|  4. THE MANAGER FACT-CHECKS      |
|  Breaks the answer into          |
|  individual claims. Verifies     |
|  each one against retrieved      |
|  evidence. Rewrites or removes   |
|  anything that can't be backed   |
|  up by your actual data.         |
+-----------------+----------------+
                  |
                  v
       Verified, Evidence-Backed
             Response

Four checkpoints. Two quality passes. Zero tolerance for fabrication.

Stage 1: The Manager Plans

Every request starts with a manager. The manager receives the question, reads it carefully, and thinks deeply about what's really being asked:

• What data will we need?
• Which systems should we check?
• What's the best approach to get a complete, accurate answer?

This planning step creates a retrieval strategy before anyone touches your data.

Stage 2: The Employee Works

With the plan in hand, the employee goes to work:

• Reaches into your connected systems
• Pulls actual data from reporting dashboards, operations platforms, and databases
• Assembles an evidence-backed answer using the information they retrieved
• Uses the specific tools assigned to their role

Stage 3: The Manager Reviews

Before anything goes out, the manager reviews the work:

• Is the answer complete?
• Does it address what was actually asked?
• Is the quality sufficient?

If the answer doesn't meet the standard, it goes back for another pass.

Stage 4: The Manager Fact-Checks

After quality approval, the manager fact-checks every individual claim:

• Supported claims pass through -- the data backs them up
• Unsupported claims get rewritten using actual evidence
• Contradicted claims get removed entirely -- if the data says otherwise, it doesn't go out

Zero Tolerance for Fabrication

Nothing leaves the desk that can't be backed up by your real data. This isn't optional -- it's built into the process. Every answer. Every time.

What You See

In the chat interface, the Work Progress drawer shows each pipeline stage in real-time with checkmarks as they complete:

1. Initializing chat
2. Analyzing complexity
3. Loading configuration
4. Selecting specialist
5. Loading advisor context
6. Searching for tools
7. Planning approach
8. Executing work
9. Validating quality
10. Refining if needed
11. Fact-checking claims
12. Response ready