Machine Learning Engineering Intern - PyTorch

Three months, one model, end to end

Most internships hand you a notebook. This one hands you a service. Over three months in Hyderabad you will take our salary prediction model from a rough PyTorch prototype to something the posting form can call in real time.

What that involves

You start by reproducing the current regression baseline and confirming its numbers yourself. Then you rebuild the training loop so it is reproducible: seeds fixed, configuration in one file, metrics logged per run. Once training is boring, the interesting part begins. You quantise the model, benchmark latency against the plain scikit-learn fallback, and decide honestly whether the deep model earns its complexity. If it does not, you write that up and we ship the simpler thing without anyone being precious about it.

Fit and terms

This suits final-year students and recent graduates who have written PyTorch by hand rather than only through a wrapper, and who can read a stack trace without panic. No professional experience is expected or asked for. You work hybrid from our Gachibowli office two days a week. The position is unpaid and deliberately structured around that: fewer hours, heavier mentoring, and documentation you can point at afterwards.