Science Academies' summer research fellowship: applying from data

Everything below comes from the SRFP 2027 announcement and the application instructions. Read both yourself before applying, because the official documents are the authority, not this summary.

What the fellowship gives you

• Two months working with a guide, at any time during 2027, preferably in summer.
• Round-trip train fare and a monthly fellowship.
• A research experience you can describe in detail when you later apply for data roles or a master's programme.

Step 1: are you eligible this year?

Eligibility depends on your exact degree and year. From the 2027 announcement:

Also check:

• [ ] A minimum 65% average in core subjects, excluding languages, from class X onward, as the announcement defines it.
• [ ] You aren't a PhD student.
• [ ] You haven't already availed the fellowship twice.

If you're in a year that isn't listed, you can't apply in this round. Note when you will be eligible and set a reminder for that year's announcement.

Step 2: the timeline, worked backwards

Step 3: prepare the documents

• Mark sheets from class X onward, uploaded as JPEG files of 700 kb or less each. Scan at a readable resolution and check each file before uploading; a blurred marksheet is avoidable.
• Your present teacher's contact details. The teacher is asked to provide a recommendation. Tell them before you submit, and tell them what you're applying for and why, so the recommendation can be specific.
• The write-up, 150–250 words, described next.

Step 4: write the statement of interest

The instructions ask for a write-up on a specific area of interest, not a general description of a subject, and not copied from anywhere. For a data student, "I am interested in data science and machine learning" is exactly the kind of general statement to avoid.

A structure that works for most data applicants:

1. 1The specific question you'd like to work on, in one sentence.
2. 2What led you to it, a course, a project or something you read.
3. 3What you already know that's relevant: methods, tools, a small project.
4. 4What you don't yet know and want to learn under a guide.
5. 5Which guides' areas fit, in general terms, from the list in the application.

A sample write-up (about 200 words)

This is an illustration of the level of specificity to aim for, not text to reuse. Write your own; copied statements are explicitly disallowed.

I would like to work on estimating quantities for small areas, such as districts, where survey samples are too small for reliable direct estimates. I first noticed this problem in a second-year statistics project, where I calculated district-level literacy rates from a public survey and found that the confidence intervals for several districts were wider than the differences I was trying to compare. I have since read about small-area estimation, where information from related areas and auxiliary data is combined with the direct estimate. I understand the idea of shrinking a noisy estimate toward a model-based one, and I have implemented a simple version in Python on simulated data, comparing its error with the direct estimator. I have not yet worked with hierarchical Bayesian models properly, or with the diagnostics needed to decide whether such a model is appropriate for real survey data. I would like to learn how to fit and check these models, and how to report their uncertainty honestly. I am interested in guides working in survey statistics, Bayesian methods or computational statistics, and would be glad to work on any related problem where estimates must be made from limited data.

Notice what it does: a narrow problem, a real (if small) prior attempt, an honest gap, and a fit to guides without naming one it hasn't spoken to.

Common mistakes

• Writing about "AI" or "data science" in general.
• Listing tools without saying what you did with them.
• Asking a teacher at the last minute, or giving a former teacher when the instructions ask for a present one.
• Uploading mark sheets above the size limit or in the wrong format.
• Submitting two applications. Only one is allowed.

If you're not selected

It's a competitive fellowship. The preparation still pays off: your statement is the first draft of a research or master's statement, and the project behind it is a portfolio piece. Our guide to presenting hackathon and Kaggle work shows how to write such projects up, and MSc, MCA, MTech or job first helps if you're weighing further study.

If research has convinced you that you want to build models on real data, the Data Science programme spends six months on statistics, machine learning and projects with review, live online with optional labs at our Noida centre.