Recommendation Systems Intern

The feed nobody scrolls

Our jobs you might like rail currently shows the same six popular listings to everyone, and its click through rate says exactly that. Fixing it is your three month project.

Where you will start

Collaborative filtering first. You build an implicit feedback matrix from saves, applications and profile views, factorise it, and see how far that alone gets you. Then content features go in, using skill and title embeddings so a brand new listing is not invisible on the day it is posted. Cold start is the genuinely interesting part of this problem and you will spend most of month two living inside it. Offline evaluation with recall at k and a proper time based split comes before any A/B test, and your mentor will ask why you chose that split.

What we expect from you

You are a final year student or a recent graduate with Python, pandas and enough linear algebra to know what a factorisation is doing. No professional experience is required. The role is fully remote across India, with two syncs a week and an end of internship demo to the product team. If your model ships, you get to watch the rail's numbers move.