Role: ML Research Intern (Applied Deep Learning)

• End-to-end owners. You own outcomes, not tasks. If something you're responsible for falls through a crack, that's on you.
• Clear, direct communicators. Bad news doesn't get better with age. You surface problems early and you write with precision.
• Uncompromising on quality. You have a quality bar that's yours, not your manager's.
• AI-obsessed. You see AI as how work gets done. If you're doing something manually that AI could handle, you feel that as friction.
• Structured thinkers. Given a vague problem, you break it down and arrive at a position instead of waiting to be told the answer.

• Build the data collection pipeline from scratch: scraping, API ingestion, deduplication, storage. Realistically this is your first two weeks and a chunk of every week after that
• Clean, label and structure raw data into training and evaluation sets that survive scrutiny
• Design the evaluation setup before the modelling starts. Baselines, metrics, splits, and what would count as a real result rather than a lucky one
• Fine-tune and adapt transformer models against the problem. Depending on where it goes, that could mean encoder models, LLM fine-tuning, PEFT/LoRA, or embedding and retrieval work
• Run experiments reproducibly. Version the data, log the runs, track the configs. An experiment you can't rerun didn't happen
• Read the surrounding literature and bring back what applies. You should be able to tell us what has already been tried and why it did or didn't work
• Write up method and results in publishable form, with support from the team on framing
• Report progress weekly in writing, including what failed. A negative result reported on Friday is useful. The same result surfaced two weeks late is expensive

• Currently in your 2nd or 3rd year of an undergraduate program (CS, math, stats, or equivalent), or a Master's student with time to commit
• Available for a continuous two-month block, either full-time or roughly 20 hours a week alongside college. Either works, but tell us which one you're signing up for
• Strong Python. Pandas, numpy and PyTorch. You should be able to write a training loop without copying one
• You have scraped something real. requests or httpx, BeautifulSoup or Scrapy, pagination, rate limits, HTML that fights back. Not a tutorial run against a clean site
• A working understanding of how transformers actually work. Attention, tokenisation, embeddings, what pretraining and fine-tuning do to a model. You should be able to explain why attention is quadratic without looking it up
• You have fine-tuned or adapted a pretrained model at least once. On your laptop, in Colab, on rented compute, we don't care where. HuggingFace transformers and datasets should be familiar
• You can read an ML paper and pull the method out of it, including the parts the authors are quietly not telling you
• Comfortable with git and working in a shared repo

• A public GitHub with real projects on it, a Kaggle record, or a paper you have already written or co-written
• You have built a dataset that didn't exist before, especially from messy or non-standard sources
• LoRA, QLoRA or other PEFT work on constrained compute
• Experiment tracking with Weights and Biases, MLflow, or a disciplined homegrown setup
• Vector databases, embedding retrieval, or RAG pipelines
• Audio, vision or multimodal data alongside text
• Any open source ML contribution, however small

• Cursor or Claude Code as your primary development environment. Hand-writing scaffolding, plots and data wrangling is time you don't have
• AI for literature work. Summarising papers, tracing citations, checking whether your idea has already been tried and published
• You read and verify everything the model produces. AI-generated code that silently mislabels your training set isn't a bug, it's a wasted month
• AI on the tedious parts so you have more time on the parts that need judgment, like experiment design and error analysis

• You want to skip the data work and only touch the modelling. Most of this internship is data work, and that is the job
• You want a structured internship with a curriculum and a daily standup. The problem is defined. The path isn't
• You can't commit a continuous block. Exams landing in the middle will break the timeline
• Your ML experience is entirely coursework and tutorial notebooks
• You expect a paper to be handed to you. Authorship follows contribution

• Duration: 2 months to start. If the work is good, we extend to 6 months
• Commitment: full-time or part-time at roughly 20 hours a week. Your call, flexible either way
• Location: remote. If you're in Bengaluru, we'd like you in the office for working sessions
• Stipend: [confirm before posting]
• What you get out of it: named authorship on the papers, a completion certificate, and a reference from the team