Human Mentor–Mentee Matching System

NLP-based matching system for pairing alumni mentors with mentees based on holistic life experience — built for the IIT Madras alumni mentoring programme.

Most matching systems — job boards, recommendation engines, dating apps — work on structured data: keywords, categories, ratings. Mentor–mentee matching in an alumni network is a harder problem. The data is unstructured (free-text descriptions of career journeys, interests, goals), the relationship is long-term, and the match quality depends on compatibility across dimensions that don’t map neatly to any single feature vector.

This project builds a matching pipeline for the IIT Madras alumni mentoring programme, developed in collaboration with a team of senior alumni. My role covers the matching algorithm and automation end of the system.

human response match app

The problem

Given a set of mentee profiles (goals, background, areas seeking guidance) and mentor profiles (expertise, experience, availability), find pairings that maximise compatibility — not just on surface-level keywords, but on the underlying substance of what each person brings and needs. No off-the-shelf solution exists for this specific formulation.

Approach

  • NLP-based profile encoding: free-text responses processed using sentence transformers to generate dense semantic embeddings
  • Similarity scoring: cosine similarity between mentee and mentor embedding vectors, weighted by configurable priority dimensions
  • Batch matching: automated assignment across the full cohort, with output rankings for human review before finalisation
  • Deployed as a Streamlit web app for programme coordinators to upload data and generate matches without touching code

Stack

Python · sentence-transformers · scikit-learn · Pandas · Streamlit. tf-idf