Admin panel
AI Rule: A Third Rule Type That Reads Free Text
Your matching algorithm can now read what both sides wrote about themselves and score how well they fit.
The matching algorithm had two rule types. Standard and custom rules compare values: same city, different department, this option against that one. People do not describe themselves with dropdowns. The paragraph a mentee writes under “what do you want help with?” sat somewhere no rule could read. The third rule type reads it. You pick one mentor field and one mentee field, and the rule reads both texts and scores how well they fit. That score joins the same matching percentage as every other rule.
Two fields, one comparison
You pick a mentor field and a mentee field. Shared ground: both sides are strong in the same area. Complementary: what the mentor is strong in is what the mentee needs.
The theme sets the axes
Each theme has a handful of axes. For corporate career they are leadership, craft depth, career moves, working across an organisation, and balance. Left on automatic, the theme is chosen from your programme’s audience.
Two sides, two different questions
On each axis the mentor is asked whether they show real experience they could pass on. The mentee is asked whether they have a goal or a need there. “I led a team” and “I want to lead a team” cannot be told apart by one question.
Jev gives the judgement
Jev is a language model that writes no text. It returns one score per question. The questions are in English, and the text is judged in whatever language it was written in.
Everyone is read once
Readings are per person, not per pair. A programme with 200 mentors and 120 mentees takes 320 readings, not 24,000 pairs. A pair’s score is calculated from those two readings.
Only the field you picked is sent
Names, e-mail addresses and contact details are not sent. Processing happens only when you run a calculation. Readings are kept, so the same text is not read twice.
The score joins the percentage, the decision stays yours
The rule takes a weight and can be made mandatory. A rule that could not be scored is not hidden: the match details say why. Nothing is matched without your approval.
What it is good for
- Reading the goal text: what a mentee writes about the help they want now counts in the matching.
- Topics that do not fit a list: useful when expertise is too varied to turn into a set of options.
- Separating the ranking: next to rules almost every pair passes, it adds a signal that tells them apart.
- Less reading by hand: you start from a ranked list instead of opening hundreds of profiles.
Where to find it Admin panel → Matching → Matching Algorithm. Access is opened per programme, and you can request it from the card.