Running AI Matching

MentorNeko's AI match engine analyzes every enrolled member's profile and generates ranked pairing recommendations. It does not make final decisions: all recommendations flow into the Shadow Approval Queue, where you review and approve them before any member is notified. This article walks you through how to configure a matching run and what to expect from the results.

How the AI Match Engine Works

When you trigger a matching run, the engine evaluates all eligible members in the selected program against one another. It considers each member's full profile across multiple dimensions, weighing them against the preferences and needs of potential partners. For each promising pair, the engine generates a rationale explaining why the match is recommended.

The output is a ranked list of pairings, from strongest to weakest match quality. You see this list in the approval queue, not the members.

What the AI Considers

The engine looks at a combination of structured profile data and compatibility signals:

• Skills and expertise: Tags that describe what each person brings and what they want to learn.
• Goals: What each member is trying to achieve in the program.
• Seniority and career stage: Whether the mentor has meaningful experience in the areas the mentee wants to grow.
• Department and industry: Relevant context for career-focused programs; may be deprioritized for cross-functional programs.
• Personality and work style dimensions: Any custom dimensions your organization has configured, such as communication preferences or coaching style.
• Overall compatibility: A holistic signal derived from the combination of all dimensions.

The relative weight each dimension carries is configurable per program. See the Configuring a Matching Run section below.

Profile Readiness Pre-Filter

Before the matching run begins, the engine automatically excludes any enrolled member whose profile is incomplete. A profile is considered ready when it has a profile photo, a bio of at least 30 words, and values set for all dimensions your organization has marked as required.

Members who are excluded due to an incomplete profile receive an automated notification prompting them to complete their profile so they can be included in the next run. They do not receive a match recommendation until their profile is complete.

This pre-filter prevents low-quality matches based on sparse data, so the results you review in the queue are based on complete information.

Bidirectional Re-Pairing Prevention

The engine automatically excludes pairs that have already been matched, regardless of how that match ended. This applies globally across all programs and uses a sorted pair key (combining both member IDs) to identify previously matched pairs.

If a mentoring relationship ended on good terms and both members want to re-pair, you can still create a manual match from the match management page. The re-pairing prevention applies only to the automated engine recommendations.

Configuring a Matching Run

Navigate to Admin > Programs, open your program, and click the Matching tab.

Step 1: Select the Program

If you have multiple programs, confirm you are on the correct one. The matching run only considers members enrolled in the selected program.

Step 2: Review the AI Prompt Template

The AI uses a prompt template to frame its evaluation of each potential pair. The default template is a good starting point for most mentorship programs. If your program has a specific focus, you can edit the template to add context.

Examples of useful additions to the prompt template:

• "Prioritize pairs where the mentor has direct experience in the mentee's stated career transition goal."
• "This is a technical skills program; deprioritize geography and department in favor of skills alignment."
• "Members have indicated whether they prefer structured or exploratory coaching styles; weight this compatibility heavily."

Avoid instructions that would cause the AI to consider protected characteristics (age, gender, ethnicity). The system is designed with PII safety guardrails.

Step 3: Adjust Dimension Weights

Each profile dimension has a weight that controls how much it influences the compatibility score. Weights are expressed as sliders from low to high.

The right weights depend on your program's purpose:

• For a leadership development program: weight goals and seniority highly.
• For a technical mentorship program: weight skills tags most heavily.
• For a peer networking program: consider weighting personality and communication style over seniority.

You can return to these settings at any time. Changes take effect on the next matching run, not retroactively on existing recommendations.

Step 4: Run the Match Engine

Click Run Matching. The engine processes in the background. For large programs, this may take a few minutes. You will receive a notification when the run is complete and results are available in the approval queue.

Reviewing Results

Once the run completes, the recommendations appear in the Shadow Approval Queue. See the Reviewing and Approving Matches article for a full walkthrough of what to do next.

No member is notified at this stage. The entire approval queue is visible only to admins.