Coimbatore-based computer vision team wants an intern who's actually trained a model end to end, not just followed a tutorial to eighty percent accuracy and called it done. This is a full-time internship, on site at the Saravanampatti office, working on image processing pipelines that feed into real products. Zero prior professional experience is fine. What isn't fine is showing up without ever having debugged a model that refused to converge.
The team builds and trains predictive models, digs through large datasets to pull out what actually matters, pushes models into production, and then keeps watching how they perform once real-world data starts hitting them instead of a clean validation set. As an intern, you'll get real ownership of pieces of that pipeline early, usually starting with data preprocessing and augmentation work in your first two weeks before moving into model training under a mentor's review.
Required skills are Python, OpenCV, at least one deep learning framework such as PyTorch or TensorFlow, and hands-on image processing experience, whether from coursework, personal projects, or a prior internship. The team specifically wants to see a project you can walk through in detail, including what didn't work the first time, since that tells them more about how you debug than a polished final result ever could on its own.
A bachelor's degree in computer science, statistics, or a related quantitative field is the baseline the company expects, and this internship is open to final-year students or recent graduates. No professional work history is required for this one; the door here is genuinely open to someone whose only prior model training happened in a classroom or a side project.
Mentorship here is structured, not incidental. Each intern gets paired with a senior engineer for the full duration, with a weekly one-on-one that's specifically about the intern's growth, separate from the daily standup about the actual work. Naukri Mitra has seen a handful of interns from this program convert to full-time roles over the past two years, and the team is explicit that this internship is treated as a real pipeline into a permanent seat, not a summer of coffee runs and shadowing.
Expect the pace to be quick. Datasets here run into the hundreds of thousands of images, and training runs on the shared GPU cluster get scheduled in blocks, so part of the job is learning to plan experiments efficiently instead of burning a slot on a run that was doomed from a bad hyperparameter choice. Nobody expects perfection on the first attempt, but repeating the exact same mistake twice tends to get noticed.
Interviews move fast too: a take-home task involving a small image dataset, followed by a technical conversation where you walk through your approach and defend the choices you made. Candidates who over-engineer the take-home task, adding complexity the problem didn't call for, tend to do worse in that follow-up conversation than those who kept it simple and could explain every decision.
The office in Saravanampatti sits inside a growing tech cluster that's picked up a fair number of data and AI teams over the past three years, largely because rents here run well below what the same square footage costs in central Coimbatore. It's a quieter setting than a big-city tech park, and most interns on the team say that's exactly what makes it easier to focus on a training run instead of getting pulled into fifteen unrelated meetings a day.
Past interns describe the hardest part not as the machine learning itself but as the discipline of writing clean, reproducible experiment code from day one, since a training pipeline nobody else can rerun is close to useless once you move on to the next dataset. The team has a lightweight internal standard for this, and you'll be expected to follow it by your second week, with code review feedback focused specifically on reproducibility as much as on model accuracy.
The role is based at Saravanampatti, Coimbatore, pin code 641035, on site five days a week. The stipend runs up to ₹21,500 a month. The team is looking to start someone within three to four weeks, in time to onboard before the next dataset refresh lands.