Intern Researcher —
Machine Learning

Internships in this country are often free coffee and photocopying. This one is neither. It is paid, it is six months long, and at the end of it you will have shipped machine learning to customers who pay us for it.

SerendAI is our applied machine-learning product. It reads the paperwork that small businesses in Sri Lanka bury themselves in, answers the questions customers keep asking, and drafts the replies staff used to stay late to write. It is not a demo. It is software that people open on a Tuesday morning and trust to do its job.

We are opening one research internship on the team that builds it. For six months, you will sit with three ML engineers, a product designer, and whichever customer has the thorniest data that week. You will run experiments that matter, write them up honestly, and — at the end of it — publish one piece under your own name on the OneCeylon notebook. A byline, not an acknowledgement.

What the six months look like

• Month 1 — the apprenticeship. You shadow, you read the codebase, you replicate one of our recent experiments from scratch and tell us where the documentation is wrong.
• Months 2 to 4 — the project. You take ownership of a single research question, scoped with your mentor. Something real — a retrieval problem, an evaluation framework, a small-model specialisation study.
• Month 5 — the writing. You draft an engineering post for the notebook. It is edited, it is argued over, and it is published under your name with a permanent URL.
• Month 6 — the handover. You ship your work into the product or its evaluation harness. You leave with something you can point to for the rest of your career.

What we are looking for

• Currently enrolled — undergraduate or early graduate — in computer science, statistics, engineering, mathematics, or a closely related field at a recognised university.
• Comfortable in Python and at least one ML framework. PyTorch is our house style, but Jax or TensorFlow is fine if you can read ours.
• Foundations in linear algebra, probability, and optimisation — the kind you actually remember, not the kind you crammed.
• Evidence you can finish something. A research project, a Kaggle notebook you are proud of, a course final that went beyond the brief. We read it properly.
• The habit of writing. If you can explain a paper to someone who has not read it, you will do well here.

Not required, but nice

• Prior exposure to NLP, retrieval, or evaluation of language models.
• An undergraduate thesis or research project in progress.
• Reading knowledge of Sinhala or Tamil — much of our data comes in these languages.
• A blog, a public repo, or any other evidence you write voluntarily.

What you get in return

• LKR 80,000 to 120,000 per month, based on experience, for the full six months.
• A mentor — one named person, one calendar invite a week, for the whole six months.
• Your own published byline on the OneCeylon notebook. Not a team post. Yours.
• A considered, written reference from your mentor when you leave. We take reference letters seriously, because we remember what it is like to need one.
• Access to the full OneCeylon book allowance and learning budget from day one.
• A realistic chance of a full-time offer at the end, if both sides want one. We do not promise conversion, but we do not hide it either: three of our last four interns are now full engineers here.

How to apply

Email careers@oneceylon.tech with the subject line OC-RES-02. Attach your CV, your transcript (unofficial is fine), and one page on a piece of ML work you have done — the problem, what you tried, what worked, and what you would do differently. Code link or notebook if you have one. We read every application, and we write back within five working days.