Manipal University Jaipur Students: Cracking the AI Internship Behavioural Interview with STAR Stories

You crack a behavioural interview for an AI internship by preparing six to eight short, true stories from your own projects and shaping each one into Situation, Task, Action and Result (STAR). Small experiences such as a messy dataset, a group-project disagreement or a model that failed are enough, as long as you can explain your own decisions and what you learned. This guide gives you a story bank worksheet to build those stories before the interview.

What behavioural questions are really checking in AI internship interviews

Technical rounds check whether you know concepts. Behavioural rounds check whether a team could work with you for a few weeks without constant supervision. For an AI intern, interviewers usually listen for these signals:

• Ownership: did you take responsibility for a piece of work, or just attend the meetings?
• Honesty about failure: can you say a model underperformed and explain why, without blaming others?
• Communication: can you explain a technical choice to someone who does not code?
• Learning speed: what did you do when you did not know a library, metric or tool?
• Judgement: did you notice data quality, bias, privacy or time-limit trade-offs?
• Collaboration: how did you handle disagreement, uneven workloads or unclear instructions?

Notice that none of these require a fancy project. A fraud-detection notebook from a course and a Kaggle practice dataset can both carry a strong story if you describe your actual decisions.

The STAR method, adapted to AI project work

STAR is a structure, not a script. Each letter has a specific job, and in AI work each one has a typical trap.

A good proportion is roughly 10% Situation, 10% Task, 60% Action and 20% Result. Most weak answers spend half their time on the setup and rush the action, which is the only part the interviewer cannot guess.

Where to find stories when you have only small AI projects

Students often say they have nothing to talk about because they have not shipped a product. Mine these sources instead:

• Course mini-projects, such as a classifier, a regression task or a recommender built for a lab.
• Group assignments where you handled data cleaning, evaluation or the presentation.
• Hackathons, club events and department competitions, including ones where you did not win.
• Self-study: a Kaggle dataset, a tutorial you extended, a small chatbot using a public API.
• Non-AI moments: organising a fest event, tutoring a junior, a part-time job. Behavioural questions rarely demand that the story be technical.
• Short online internships and structured tasks, where you can describe how you managed a tight deadline.

If you want a structured, deadline-driven experience to draw from, the free two-week online internship from CareerSwami gives you a 14-day window, an offer letter and a completion certificate after successful completion, and it offers 26 tracks. Whatever you choose, only tell stories about work you actually did.

The story bank worksheet: map each AI experience to the questions it can answer

Build a table like the one below in a notes app or spreadsheet. Each row is one story. The aim is that every common behavioural question can be answered by at least two different rows, so you are never forced to reuse a story awkwardly.

Question-to-story matching grid

Once your rows are filled, match them to the questions you are most likely to hear.

A worked example: turning a small project into a STAR answer

The following is an illustrative example written for teaching. It is not a real student’s story, and no code or results here were actually run.

Question: “Tell me about a time a project did not go as planned.”

Raw material in the story bank: A three-person course project to predict whether a customer would leave a telecom service. The first model showed very high accuracy, but the student noticed it predicted almost everyone would stay.

Weak version: “We built a churn model using logistic regression and random forest and got good accuracy, then we improved it.”

STAR version:

• Situation: “In my third-semester machine learning course, my team of three built a customer churn predictor for a public dataset.”
• Task: “I was responsible for model evaluation and for presenting results to our faculty reviewer.”
• Action: “Our first model reported 85% accuracy, but when I checked the confusion matrix, it was predicting ‘no churn’ for nearly everyone. The classes were imbalanced, so accuracy was misleading. I explained this to my teammates, we switched our main metric to recall for the churn class, tried class weights and resampling, and I documented each experiment in a shared sheet so we could compare them.”
• Result: “Recall on the churn class went from near zero to a level we could defend in the review, and the reviewer specifically asked about our metric choice. I now check class balance before I trust any headline accuracy.”

Why it works: it is honest about the failure, shows a specific decision, names a lesson, and includes a detail an interviewer can probe. Replace the numbers with your own real ones, and never round up.

Writing the Action section so it sounds like you, not the team

The most common weakness in student answers is the invisible “I”. Use this checklist while drafting each Action paragraph.

1. Write three verbs that describe only what you did (cleaned, compared, questioned, wrote, presented).
2. Add one “because” sentence that explains a decision.
3. Mention one alternative you rejected and why.
4. Name one person you communicated with and what you told them.
5. Cut tool lists to the two tools that actually mattered.

Handling follow-up probes

Interviewers often follow with “What would you do differently?” or “How did you know it worked?” Add two lines to every story row in your worksheet: a “do differently” line and a “how I measured it” line. These turn a rehearsed answer into a conversation.

Honest results: what to say when you have no impressive numbers

Many small projects do not have a business metric. You still have options that stay truthful:

• A technical metric with context (“F1 improved from 0.52 to 0.67 on our held-out split”).
• A time or effort change (“cut manual labelling time by reusing a script”).
• A qualitative outcome (“our reviewer used our notebook as the reference example”), only if true.
• A learning outcome (“I now start with a baseline model before anything complex”).

If you cannot recall a number, say so. “I don’t remember the exact figure, but it was a clear improvement over the baseline” is far safer than an invented statistic that a follow-up question can expose.

Manipal University Jaipur students: preparing STAR stories around your internship and project calendar

Your own course structure supplies useful story material. The university publishes semester-wise course lists for programmes including B.Tech CSE (AI & ML) and CSE (Data Science), so you can see which labs and projects are likely to sit in your bank, and a Data Science programme page lists courses semester by semester. Treat these as prompts for finding stories, not as proof of what any employer expects.

Timing matters too. The Academic Calendar 2026-2027 shows an odd semester from July to December 2026, an even semester from January to July 2027 and a marked summer break, and it says the university may modify it. A 14-day internship is easiest to fit into a break, so check the current calendar before committing.

On internship rules, be careful. A university blog says B.Tech students do summer internships and a six-month industry training in the last semester (blog text, not a regulation). The only detailed rules available are in the faculty-specific FoMCA Academic Handbook 2022-23, which is older and covers management, commerce and hospitality. It describes logbooks, mentor appraisals and S/US grading, which is a reminder that documented, specific records make interview stories easier to tell. It does not tell AI or B.Tech students how a short online internship is treated.

For interview practice, the Training and Placement Cell is described as running counselling and training sessions for placement preparedness.

Verification steps for MUJ students before you rely on an internship

1. Ask your programme coordinator in writing which internship or training rules apply to your batch and scheme year.
2. Ask whether a short online internship can count toward any requirement, and what proof (certificate, logbook, mentor letter) is accepted.
3. Check the current Academic Calendar for your semester’s exact break dates before planning the 14 days.
4. Email the placement office at the listed placement contact and ask whether mock behavioural interviews are available for your branch.
5. Keep a dated log of your work from day one so your stories have facts behind them.

A 14-day plan to build and rehearse your story bank

Delivery habits that make a rehearsed story sound natural

• Memorise the beats, not the sentences.
• Start with a one-line headline: “This is a time I caught a misleading metric.”
• Keep each answer to about 90 seconds to two minutes.
• Pause after the Result and let the interviewer respond.
• Use plain words for non-technical interviewers; save jargon for when they ask.
• If your mind blanks, ask for a moment. It reads as composure.

Frequently Asked Questions

How many STAR stories do I need for an AI internship interview?

Six to eight is a practical range. That is enough to cover failure, teamwork, deadlines, communication, learning and ethics, with backups, without forcing you to memorise too much.

Can I use a college group project if my contribution was small?

Yes, but describe only your real part and be precise about it. A small, honestly described contribution, such as evaluating the model or cleaning one table, is better than a vague claim of leading the whole effort.

What if my AI project failed completely?

That can be your strongest story. Explain what you expected, what the evidence showed, what you changed, and what you would check earlier next time. Interviewers value honest diagnosis.

Should I mention accuracy numbers in behavioural answers?

Include one or two if you remember them accurately and can explain the metric. If the interviewer is non-technical, translate the number into plain meaning, such as “it caught most of the customers who left.”

How do I answer “Why do you want this AI internship?” with STAR?

This question is usually motivational rather than situational, so STAR is optional. Link a specific past experience to what the internship offers, then state one concrete thing you want to learn.

Can a short two-week internship give me a credible story?

Yes, if you can describe what you were asked to do, how you managed the 14 days and what you produced. For a structured option, the free internship at CareerSwami runs for two weeks across 26 tracks, but your interview story should rest on your own actual work.

How do I avoid sounding rehearsed?

Practise the structure and the key facts, not exact wording. Record yourself, vary the phrasing on each attempt and let a friend interrupt with follow-up questions.

What should I do if I blank on a behavioural question?

Take a breath and say you want a moment to think of the right example. Then pick the story from your matching grid that is closest, and focus on the Action, since that part is the most persuasive.

Related Artificial Intelligence guides

• Dr. Homi Bhabha State University Students: Applying AI to Science Data in a Short Internship
• M.V.N. University Students: Evaluating an AI Internship Certificate's Worth Beyond the Paper
• IITRAM Students: AI Internship Ideas Linking Machine Learning to Transport and Infrastructure Data
• All Artificial Intelligence internship guides