Machine Learning Data Intern (scikit-learn)

Ranking is the least glamorous and most valuable machine learning problem in a job marketplace. You will spend six months on the data that makes ours work.

Where you fit

You sit with two data scientists in Bengaluru, three days a week in the office. Your remit is the training data rather than the model architecture: assembling labelled examples of good and bad candidate to job matches, auditing them for leakage, and building the offline evaluation harness that tells us whether a model change is genuinely an improvement.

Specific things on the list

Label a stratified sample of ten thousand applications against recruiter outcomes. Build resume skill extraction as a scikit-learn pipeline so training and serving share one code path. Implement precision at k and mean reciprocal rank, then run the current production ranker through them to establish an honest baseline.

Being useful here

Half of applied machine learning is refusing to trust a result that looks too good. If you enjoy digging into why validation accuracy suddenly jumped, and you have the Python fundamentals from coursework or personal projects, the rest is teachable. No industry experience needed, only evidence that you finish what you start.