A data analyst portfolio is three or four finished projects, each answering one business question with a real dataset, documented well enough that a stranger can follow what you did. Not twelve half-built notebooks. Three, done properly, beats twelve every time.
The reason most portfolios fail is not technical. It is that they show a process without a point. A notebook full of charts with no stated question and no conclusion tells a hiring manager nothing about whether you can be useful on a Monday morning.
What a hiring manager actually opens
Assume ninety seconds of attention. In that time the reviewer looks for a readable summary at the top of your repository, a chart or two that make sense without explanation, and evidence that you handled messy data rather than a pre-cleaned teaching file.
What they are checking, in order: can you frame a question, can you get data into a usable state, can you reach a defensible answer, and can you explain it to someone who is not an analyst. The last one carries far more weight than beginners expect. Plenty of candidates can produce a correlation matrix. Very few can say what a business should do about it.
The four parts of one good project
| Part | What it contains | What the reviewer checks |
|---|---|---|
| The question | One specific business question, written in a sentence | Whether it is answerable and worth answering |
| The data | A named public dataset, with its source and date | Whether it is real and whether you understood its limits |
| The analysis | Cleaning steps, the logic, the intermediate checks | Whether your conclusion survives scrutiny |
| The result | Two or three charts and a short written answer | Whether you can communicate to a non-analyst |
Write the question first, before you open the file. “Which Bengaluru neighbourhoods offer the best-rated restaurants under ₹800 for two” is a project. “Restaurant data analysis” is not.
Indian datasets you can download today
Every dataset below is public, free, and specific enough to build a real project on. Using Indian data has a practical advantage in an Indian interview: the reviewer knows the domain and can tell whether your conclusion sounds right.
| Dataset | Where to get it | What is inside | A question it can answer |
|---|---|---|---|
| Open Government Data Platform | data.gov.in | Thousands of central and state datasets covering crop prices, rainfall, road accidents, school enrolment | Which states cut road accident deaths fastest, and did highway spending track that |
| RBI Database on Indian Economy | data.rbi.org.in | Banking, credit, deposits, inflation, exchange rates and state finances by year | How bank credit growth differs between southern and northern states |
| Ministry of Statistics | mospi.gov.in | Consumer Price Index series, National Sample Survey rounds, GDP releases | How food inflation diverged from headline CPI over the last five years |
| Census of India | censusindia.gov.in | District-level population, literacy, housing and migration | Which districts sit far above their own state average on female literacy |
| Telangana Open Data | data.telangana.gov.in | State datasets on health, transport, agriculture and budgets | Does district health spending track measurable health outcomes |
| CPCB air quality | airquality.cpcb.gov.in | Hourly pollutant readings from monitoring stations across Indian cities | How Bengaluru’s pollution pattern differs from Delhi’s hour by hour |
| Vahan dashboard | vahan.parivahan.gov.in | Vehicle registrations by state, fuel type and vehicle class | Where electric two-wheeler adoption is growing fastest, and where it stalled |
| eNAM mandi prices | enam.gov.in | Daily arrivals and prices for agricultural commodities by mandi | How far onion prices diverge between mandis in the same week |
| Zomato Bengaluru restaurants | Kaggle | Around 50,000 restaurant listings with cuisine, cost for two, ratings and locality | Which localities give the best rating per rupee |
| IPL ball-by-ball data | Kaggle | Every delivery from every IPL match, with venue, players and outcome | Whether winning the toss changes the result at a specific ground |
Open City also aggregates civic datasets for Indian cities at opencity.in, which is useful when you want something local to where you live.
Four project briefs you can start this week
Bengaluru restaurant pricing. Take the Zomato Bengaluru dataset. Clean the cost column, which arrives as text with commas. Group by locality and cuisine, filter to places with enough ratings to be meaningful, and find where rating per rupee is highest. The deliverable is a ranked table of localities plus a scatter chart of cost against rating, and a paragraph naming three localities worth opening a mid-price restaurant in. Tools: Excel or pandas, then a chart.
Air quality across two cities. Pull a year of hourly CPCB readings for one Delhi station and one Bengaluru station. Handle the missing hours honestly rather than deleting them silently. Build a monthly average, then an hour-of-day profile. The deliverable is a line chart per city and a written answer on when each city’s air is worst and why the shapes differ. Tools: pandas, Matplotlib.
Electric vehicle adoption by state. Use the Vahan dashboard for two-wheeler registrations split by fuel type. Calculate the electric share per state per year, then rank states by growth rather than by absolute volume, because absolute volume just re-ranks states by population. The deliverable is a bar chart of electric share by state and a note on which state subsidies coincide with the jumps. Tools: Excel or pandas.
Mandi price spread. Take eNAM daily prices for one commodity across ten mandis for a season. Compute the spread between the highest and lowest mandi each day, and find whether the spread widens at harvest. The deliverable is a chart of the spread over time and a short answer on where a trader would have made money. Tools: SQL or pandas.
Each of these takes a weekend or two. None of them requires machine learning, and none of them should. Analyst roles are hired on analysis.
Projects to avoid
The Titanic survival classifier. It has been done tens of thousands of times, it is a modelling exercise rather than an analysis, and reviewers stop reading at the word Titanic.
The Iris flower dataset. Same problem, plus 150 rows is not data.
Any tutorial you followed step by step. If the reviewer can find the identical notebook on YouTube, the project shows only that you can type. Change the question, change the dataset, or both.
A dashboard with no question. Twenty charts on one screen is not a portfolio piece. One chart that answers something is.
Where to put it
Use GitHub as the home for the work and a public link for the visual. A repository per project, not one repository called “projects” holding everything.
Inside each repository, keep it plain: a README, a data folder or a link to the source if the file is large, a notebook or script, and an outputs folder with the charts as images. The README is the single most important file, and it needs five things in this order: the question, the dataset and its source with a date, what you did to clean it, the answer, and the caveats. Write it for a manager, not for another analyst.
If your project is visual, publish the dashboard to Tableau Public or Power BI’s public sharing and link it from the README, so the reviewer sees the result without cloning anything. Then write one LinkedIn post per project summarising the finding in four lines with one chart. Several SkilloVilla alumni describe that posting habit as the thing that started getting recruiters to approach them rather than the other way round.
How to talk about it in an interview
Expect the question “walk me through a project” in almost every first round. Answer in this shape: the question, why it mattered, one specific problem in the data and how you handled it, the answer, and what you would do differently with more time.
That last part is not a trap. Naming a limitation in your own work is the fastest way to sound like someone who has done this before. Say the CPCB station had four weeks of missing readings and you excluded rather than interpolated them, and explain why. That single sentence does more than an extra chart. If you want the fuller picture of what recruiters check beyond projects, our guide to data analyst qualifications in India covers the rest, and what placement assistance actually means is worth reading before you rely on any programme to do the job hunt for you.
Frequently asked questions
How many projects should a data analyst portfolio have?
Three to four finished projects is the right number for a fresher. Each should use a different skill combination, so one heavy on SQL, one on cleaning messy data with Python, and one that ends in a dashboard. Beyond four, reviewers do not read further, and adding more usually means each one gets less attention than it needed.
What projects should a beginner data analyst do?
Pick a dataset from a domain you can reason about, then ask a question a business would pay to answer. Retail pricing, vehicle registrations, mandi prices, air quality and cricket data all work well because you can sanity-check the result against what you already know. Avoid classification exercises on toy datasets, since they test modelling rather than analysis.
Do I need GitHub for a data analyst portfolio?
Yes, or something equivalent that gives a public link. GitHub is the default, it costs nothing, and recruiters know how to read it. If your work is dashboard-heavy, pair it with Tableau Public or a published Power BI report so the visual is one click away rather than behind a download.
Can I build a data analyst portfolio without work experience?
That is exactly what a portfolio is for. Public datasets remove the need for a company to hand you data, and a well-documented project with a clear question is treated as evidence of capability by most Indian hiring managers for junior roles. Several alumni featured on this blog moved into analyst jobs from civil engineering, HR and sales on the strength of project work.
Should my portfolio projects use machine learning?
Not for an analyst role. Hiring managers are assessing whether you can clean data, ask a sensible question and communicate an answer, and a model tends to obscure all three. Add a simple regression only if it genuinely answers the question you posed, and be able to explain every assumption behind it.
How long should each portfolio project take?
Roughly one to two weekends each once you know the tools. If a project drags past a month, the question was probably too broad, so narrow it to one city, one commodity or one season. A finished narrow project is worth far more than an abandoned ambitious one.
Build projects with someone reviewing them
The gap between a project that looks fine to you and one that survives a hiring manager’s questions is usually invisible from the inside. That is what feedback fixes.
SkilloVilla’s Data Analytics with Python track runs 4 to 5 months of live classes at ₹71,999, currently ₹58,999, with capstone projects reviewed by working analysts and mentors from companies including Swiggy, Razorpay, Amazon and Cred. If you want the data science version with machine learning included, the Data Science & AI track covers it at ₹1,01,999, currently ₹89,999.
Fees and ratings last checked August 2026; confirm current numbers with the provider before enrolling.
