Deep Learning Research Internship

An open question, not a ticket

Can a model tell, from the text of a job post alone, whether the compensation it advertises is plausible for that role in that city? Nobody here knows, and over three months you will find out.

Method

You assemble a dataset of Indian internship posts, build a text regression baseline in scikit-learn using TF IDF, then move to a fine tuned transformer from Hugging Face and test whether the extra capacity earns its keep. You report mean absolute error broken out by city and by domain rather than as one headline number, because the aggregate hides exactly the failure mode you care about. Ablations, multiple seeds and an honest limitations section are all expected. If the work holds up, it becomes a trust signal on the live site.

Support and setting

You work from the Mumbai office three days a week, with GPU access and a weekly one to one with the research lead. There is a reading group on Wednesdays, and you will present a paper at least once. Final year students and recent graduates are welcome. Coursework level deep learning and a willingness to be wrong in public are the real requirements here.