Gokhale Institute Students: Applying AI to Economics Data in a Short Internship

Yes, an economics student can apply AI to economics data in a 14-day internship, provided the goal is small: one clean dataset, one well-chosen simple method, and one honest check of the result. The worksheet below pairs common economics questions with beginner-friendly methods such as forecasting, clustering and classification, and lists what to verify before you trust any output.

Why economics students have an advantage in applied AI

Most introductory machine learning (ML) material assumes a computer science background. Economics students bring something that matters just as much: they know what the numbers mean. You already understand inflation versus price levels, why a regional average can hide wide spreads, and why correlation is not causation. These habits are exactly what stops a model from producing confident nonsense.

The gap is usually tooling and vocabulary. That is fixable in two weeks if you narrow the scope. You do not need deep learning. For most economics questions, a regression, a simple time-series forecast or a clustering run is more than enough, and easier to defend in a viva or a report.

What a 14-day plan can realistically cover

Two weeks is enough for one small, finished project. It is not enough for a thesis-grade study. Aim for a short report, a notebook or spreadsheet, and a one-page summary of what the model can and cannot say. The free two-week online internship from CareerSwami offers 26 tracks, an offer letter and a completion certificate after successful completion, so you can pick an AI track and use this worksheet as your project frame.

1. Days 1-2: Pick one question and one public dataset. Write the question in a single sentence.
2. Days 3-4: Clean the data. Check units, missing values, date formats and definitions.
3. Days 5-6: Build a baseline: a mean, a last-value forecast or a simple linear regression.
4. Days 7-9: Apply the chosen ML method from the worksheet.
5. Days 10-11: Validate: hold-out testing, sensitivity checks, comparison with the baseline.
6. Days 12-13: Write the report and make two or three clear charts.
7. Day 14: Review for overclaiming, tidy the files and prepare a short summary.

The worksheet: economics questions matched to simple ML methods

Use this table to choose a method. Each row names the question type, a suitable method, an example data type and what to verify. The data examples are illustrative categories, not claims about any specific dataset’s contents.

How to choose between methods

• If your question asks “what next?”, it is a forecasting problem.
• If it asks “which belong together?”, it is a clustering problem.
• If it asks “which category does this fall in?”, it is classification.
• If it asks “what is associated with this outcome?”, start with regression.

When two methods fit, choose the one you can explain in plain language to a non-technical reader.

Verification checklist for any model output

This checklist is the heart of the worksheet. Fill it in before writing conclusions.

1. Source: Who published the data, and on what date? Note the table name and version.
2. Definitions: Is the variable nominal or real? Annual or monthly? Fiscal or calendar year?
3. Coverage: Which regions, years and groups are included or missing?
4. Baseline: What does the simplest possible prediction score?
5. Split: Did you test on data the model never saw? For time series, was the test period later than the training period?
6. Stability: Do results change a lot when you drop a year or change a parameter?
7. Plausibility: Would an economist find the pattern sensible? If not, investigate before reporting.
8. Claim strength: Does your wording say “associated with” and “predicts”, not “causes”?

Worked example: clustering states by economic profile (illustrative)

This example is illustrative. No code was run and the numbers below are invented to show the steps, not real data.

Question: Can states be grouped by similar economic structure to compare policy needs?

1. Data: Build a table with one row per state and columns such as per-capita income, share of agriculture in output, share of services, literacy rate and urbanisation. Record each source and year.
2. Clean: Confirm all columns refer to the same or nearby years. Note any state with missing values and decide whether to drop or fill it, writing down the decision.
3. Scale: Standardise each column so income in rupees does not dominate percentages.
4. Method: Run k-means with 2 to 6 clusters. Compare them using the silhouette score and, more importantly, whether each cluster tells a coherent story.
5. Interpret: Suppose three clusters emerge: a high-income, service-heavy group; a mid-income mixed group; and a lower-income, agriculture-heavy group. Name them cautiously and list the average of each variable.
6. Verify: Rerun without one variable. If many states switch clusters, the grouping is fragile and should be reported as such.
7. Report: State that clusters describe similarity in the chosen indicators, not causes of development.

A short sample of what a notebook outline might look like (illustrative):

1. Load state-level table
2. Check missing values and years
3. Standardise numeric columns
4. Fit k-means for k = 2..6
5. Compare silhouette scores
6. Profile clusters by variable means
7. Sensitivity: drop one variable, refit, compare

Common data traps in economics datasets

• Base-year changes: Price and output series are sometimes rebased or revised, which can create fake jumps when old and new series are joined.
• Nominal versus real values: Forecasting nominal GDP without adjusting for prices mixes growth with inflation.
• Small samples: Annual state data gives few observations, so complex models overfit quickly.
• Survey weights: Household surveys often need weights to represent the population. Ignoring them can distort averages.
• Look-ahead leakage: Using a variable that would not have been known at prediction time inflates accuracy.
• Shocks: Events such as the pandemic can break historical patterns, so a model trained across them may behave poorly.

Gokhale Institute students: fitting an AI-on-economics-data project into your programme

GIPE describes itself as an institution given Deemed to be University status in 1993. Its 2026-27 admission brochure lists economics-based programmes only, with no MBA or computer science programme. That makes a project that applies ML to economic data a natural fit: your training is in the subject, and the project adds a technical layer.

If you are in the B.Sc. Economics (Data Science) programme, the same brochure says it covers R and Python, database management systems, machine learning, AI and Generative AI, so this worksheet can reinforce what you study. Students in other programmes may be learning these tools for the first time, so keep the method simple.

The brochure also says internships are part of the evaluation system for First Year PG and Third Year UG students, and that the M.A./M.Sc. structure includes “OJT (On Job Training) / Internship” and “Research Project Economics” as compulsory components. It describes an 8-week summer internship in May-June. A 14-day online internship is therefore shorter than the institute’s described summer internship. The pages reviewed did not state credits, semester placement or whether an online or self-arranged short internship is accepted, so do not assume it counts towards any requirement. Treat it as skill-building and portfolio work unless the institute confirms otherwise in writing.

The brochure also mentions a faculty Placement and Internship Committee guiding a student-run implementation cell, and a placements committee page that, as an excerpt, lists a Faculty In-Charge for Placement and Internships. Check the current version of that page for names and contacts.

Verification steps for GIPE students

1. Email the Placement and Internship cell, copying your programme coordinator, and ask whether a 14-day online internship can be recorded in any form for your programme.
2. Ask in writing whether your programme’s internship or OJT requirement has a minimum duration, and request the current regulation or circular.
3. Ask whether an internship certificate must come from a particular type of organisation or carry specific details.
4. If you plan to use the project for your research project, ask your guide whether data sources and methods need prior approval.
5. Check the current website for the 2026-27 academic calendar and plan the 14 days around exams and term work.
6. Keep the offer letter, certificate and project files together in case a faculty member asks to see them.

Packaging the project so it reads well to a faculty reviewer or recruiter

• One-page summary: Question, data, method, result, limitation.
• Reproducible notebook or spreadsheet: Someone else should be able to follow your steps.
• Two or three charts: A forecast with its error range, a cluster profile, or a coefficient plot with intervals.
• Limitations paragraph: Name what the model cannot tell you.
• Plain-language takeaway: One sentence a policy reader could use.

Frequently Asked Questions

Do I need to know programming to try this worksheet?

Basic familiarity helps, but you can start with spreadsheet forecasting and then move to Python or R. If you have used R in an econometrics course, you already have a head start for regression and clustering.

Which ML method should an economics beginner learn first?

Start with regression and a simple time-series forecast. They connect directly to econometrics, and they give you a baseline against which any fancier method must prove itself.

How do I know if my forecast is any good?

Compare it with a naive baseline, such as repeating the last observed value, on a recent period the model has not seen. If it does not beat that baseline, the added complexity is not helping.

Can clustering tell me why regions differ?

No. Clustering shows which regions look alike on the chosen indicators. Explaining why requires economic reasoning and often a different kind of analysis.

Where can I find data for a short project?

Use official statistical releases and public data portals, and record the table name, publication date and definitions. Prefer datasets that are small enough to clean within two or three days.

Is it a problem if my model finds a pattern that contradicts economic theory?

Not necessarily, but investigate first. Check for data errors, a structural break or a leaked variable. If the result survives these checks, report it with caution and describe the limits.

How should I describe this project on a CV?

State the question, dataset, method and one result, for example “Clustered states on five economic indicators and tested stability by dropping variables”. Avoid claiming causal findings or policy impact.

Which CareerSwami track suits an economics student wanting to apply AI to data?

Look among the 26 free two-week tracks for an AI or data-oriented one, then check that its tasks let you work on a dataset of your own choosing. Confirm the track description before enrolling so the worksheet maps onto it.