New opportunity

AI/ML Engineer

SetuAI · Pune, India

About this role

SetuAI seeks an AI/ML Engineer to build production recommendation systems and adaptive algorithms that personalize learning paths for students. The role uses data from more than 50,000 students to improve content recommendations, difficulty adaptation and predictive performance models. It is an applied ML role rather than a research position.

Responsibilities

Design and improve adaptive learning algorithms that personalize content for each student.
Build and iterate on recommendation models for question selection, topic ordering and revision scheduling.
Analyze student performance data to identify patterns and inform product decisions; work with the product team to turn learning science principles into features.
Establish ML pipelines, model evaluation frameworks and A/B testing infrastructure. Monitor production models and respond to regressions.

Required qualifications

At least 3 years of ML/AI engineering experience in a production environment.
Strong Python skills, including pandas and scikit-learn, plus PyTorch or TensorFlow.
Experience with recommendation systems or collaborative filtering; proficiency with SQL and large structured datasets.
Understanding of ML evaluation metrics and A/B testing methodology.

Preferred qualifications

Experience in EdTech, knowledge tracing or cognitive science applications; familiarity with LLMs used in tutoring or Q&A systems; or a background in Item Response Theory or Bayesian Knowledge Tracing.

The full-time position has a hybrid setup in Pune with flexible remote days. Benefits listed include competitive salary and ESOPs, a ₹30,000-per-year learning stipend, health insurance and wellness benefits.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

Pythonpandasscikit-learnPyTorchTensorFlowSQLRecommendation systemsCollaborative filteringAdaptive learningMachine learning pipelinesModel evaluationA/B testingData analysisKnowledge tracingLarge language modelsItem Response TheoryBayesian Knowledge Tracing

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Personalize learning paths
SetuAI says this role designs and improves adaptive algorithms, including difficulty adaptation, using data from more than 50,000 students.
Improve learning recommendations
The role builds and iterates on models for question selection, topic ordering, and revision scheduling. Recommendation-system or collaborative-filtering experience is required.
Turn performance data into product features
The posting calls for analyzing student performance patterns and working with product to translate learning science principles into engineering features.
Ship and evaluate production ML
Responsibilities include ML pipelines, evaluation frameworks, A/B testing infrastructure, production monitoring, and responses to model regressions.

Likely challenges to think through

Personalization with uneven student histories
Preparation hypothesis, not a stated SetuAI problem: newer students may have less performance data, creating a trade-off between individualized recommendations and reliable defaults. Prepare to explain how you would evaluate both groups separately.
Measuring learning rather than clicks
Preparation hypothesis: a recommendation that increases immediate activity may not improve later performance. Prepare to distinguish offline model metrics from A/B test outcomes and discuss how you would choose meaningful learning measures.
Iterating without destabilizing production
Preparation hypothesis: improving adaptive models while serving students requires balancing faster releases against regression risk. Prepare to describe monitoring signals, rollout checks, and a response when model performance declines.

Practice your answers

Technical practice

How would you design a system that recommends a student's next question or topic?

What to coverThe role includes question selection and topic ordering. Suggested preparation: explain how you would represent student progress and content, choose a recommendation approach, and balance learning value with variety.

How would you adapt question difficulty as a student progresses?

What to coverDifficulty adaptation is a stated focus. Suggested preparation: discuss signals of mastery, how quickly difficulty should change, and how you would avoid overreacting to a small number of answers.

How would you decide when a student should revisit a topic?

What to coverThe role includes revision scheduling. Suggested preparation: describe how you would use prior performance and time since practice, then evaluate whether the schedule improves later performance.

When would collaborative filtering help with content recommendations, and when might it fail?

What to coverRecommendation systems or collaborative filtering experience is requested. Suggested preparation: compare shared patterns across students with individual learning needs, including sparse histories and new content.

How would you build a model to predict future student performance?

What to coverPredictive performance models are part of the work; knowledge tracing and Item Response Theory are listed as optional backgrounds. Suggested preparation: define a prediction target, explain which information is available at prediction time, and compare a simple baseline with a more adaptive approach.

How would you prepare and validate a large student-performance dataset for modeling?

What to coverThe posting calls for SQL, large structured datasets, pandas, and student-performance analysis. Suggested preparation: explain joins and data-quality checks, and how you would prevent later student activity from leaking into earlier predictions.

How would you structure a reproducible pipeline for iterating on recommendation models?

What to coverSetting up ML pipelines is a stated responsibility, and the role names scikit-learn plus PyTorch or TensorFlow. Suggested preparation: outline data preparation, training, versioning, and deployment checks, including when a simpler model is preferable.

How would you evaluate recommendations before releasing them?

What to coverThe role calls for model evaluation frameworks. Suggested preparation: choose offline measures tied to question selection or learning progress, explain how you would split sequential student data, and identify what offline results cannot establish.

How would you test whether a new learning-path algorithm improves outcomes?

What to coverA/B testing infrastructure and methodology are explicitly mentioned. Suggested preparation: define the comparison, primary outcome, guardrail measures, and analysis approach while considering how repeated activity from the same student affects interpretation.

How would you detect and respond to a production-model regression?

What to coverMonitoring production models and responding to regressions are stated duties. Suggested preparation: distinguish data or pipeline failures from genuine performance declines, and explain investigation, rollback, and re-evaluation trade-offs.

Behavioural practice

Tell me about a time you translated a product goal into an ML feature.

What to coverThe role involves working with product colleagues to turn learning-science principles into features. Suggested preparation: use a real situation, your specific actions in resolving ambiguity, and the resulting product or measurement outcome.

Describe a time you found a meaningful pattern in user data.

What to coverAnalyzing student-performance data to inform product decisions is part of the role. Suggested preparation: give a real example of the question, the checks you used to challenge your interpretation, and what decision resulted.

Tell me about a production model regression you handled.

What to coverThe posting explicitly calls for monitoring and responding to regressions. Suggested preparation: explain the situation, your investigation and communication, and the outcome, without claiming the fix worked unless you measured it.

Describe a time an experiment challenged your team's expectations.

What to coverThe role includes model evaluation and A/B testing. Suggested preparation: recount a real test, how you handled unexpected or inconclusive evidence, and what action followed.

Tell me about a time you shipped an ML improvement under practical constraints.

What to coverSetuAI describes this as applied ML whose results ship to production. Suggested preparation: use a real example to explain the constraints, the trade-offs you made, and the observed result.

Sharpen your knowledge

Topics to prepare

Recommendation and collaborative filtering concepts
The posting requires recommendation-system or collaborative-filtering experience for question selection, topic ordering, and revision scheduling. Revise how these approaches rank options; as an exercise, explain when student–question interaction data would support collaborative filtering and when it would not.
Adaptive learning and difficulty
The role explicitly covers personalized content and difficulty adaptation. Sketch how a student's recent answers could change the next question's difficulty, and explain what evidence would indicate the adjustment helped.
Model evaluation
SetuAI asks for ML evaluation frameworks and knowledge of evaluation metrics. For a hypothetical question recommender and performance predictor, choose appropriate metrics and explain what each would miss.
A/B testing methodology
The posting calls for A/B testing infrastructure and methodology. Outline a test of a revised topic-ordering model, specifying the comparison, outcome measure, and how you would interpret an inconclusive result.
Production monitoring
Monitoring models and responding to regressions are stated responsibilities. Define signals you would check after a recommendation-model update and describe how you would investigate a drop in performance.
Knowledge tracing and learning science
Knowledge tracing, Item Response Theory, and Bayesian Knowledge Tracing are listed as nice-to-haves, not requirements. As optional preparation, compare what IRT and Bayesian Knowledge Tracing represent, then suggest how either might inform question selection.

Skills to revise

SQL analysis of student performance
The role requires SQL with large structured datasets and analysis of student performance. Practice writing queries that aggregate hypothetical student–question attempts by topic and time, then identify a pattern worth checking with the product team.
Python data preparation and modeling
Strong Python skills, including pandas, scikit-learn, and PyTorch or TensorFlow, are required. On a sample interaction table, prepare features with pandas and compare a scikit-learn baseline with a small model in either named deep-learning framework.
Recommendation-model iteration
Building and iterating on recommendations is a stated responsibility. Using hypothetical student–question interactions, implement a simple collaborative-filtering baseline, rank candidate questions, and inspect examples where its selections look unsuitable.
Reproducible evaluation
The posting calls for model evaluation frameworks. Practice creating a repeatable comparison of two question-selection models, recording the data split, metrics, and results so that a proposed change can be assessed consistently.
Pipeline and regression response
SetuAI explicitly calls for ML pipelines and production regression response. Sketch a pipeline from student-performance data to updated recommendations, then rehearse diagnosing a hypothetical post-update metric drop; this is a preparation exercise, not a description of SetuAI's implementation.

Your preparation checklist

Prepare a production ML project walkthrough
The role requires production ML experience. Prepare a concise walkthrough of a system you worked on, covering your contribution, data, model choices, deployment, monitoring and measurable outcomes.
Show recommendation-system evidence
The role calls for recommendation systems or collaborative filtering. Prepare an example explaining candidate generation or ranking, how you handled sparse interactions, and how you evaluated recommendation quality.
Practice an adaptive-learning design exercise
SetuAI builds question selection, topic ordering, revision scheduling and difficulty adaptation. As interview practice, sketch a learning-path design with inputs, decision logic, feedback signals and trade-offs; label it as a proposal, not SetuAI’s implementation.
Prepare a data-analysis example
The role involves student-performance analysis and large structured datasets. Bring a brief example showing a SQL query or analysis approach, the use of Python tools such as pandas, and how a finding informed a product decision.
Outline evaluation and experimentation
The role specifies model evaluation and A/B testing. Prepare a one-page plan with offline metrics, an experiment hypothesis, success and guardrail measures, and how you would interpret conflicting results.
Explain production operations
SetuAI asks this engineer to establish ML pipelines and respond to model regressions. Prepare a diagram or walkthrough covering data validation, retraining, deployment checks, monitoring and a response to degraded performance.
Draft focused questions for the employer
Ask how the team currently measures learning-path quality, where recommendations most need improvement, and how product and ML engineers turn learning-science ideas into features. These are questions to investigate, not assumptions about existing processes.

Availability last checked: 5 Nov 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in India. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…