Glean AI India Machine Learning Engineer Interview: Rounds, Questions and Preparation

What the job involves

Machine Learning Engineers at AI and deeptech companies like Glean AI India build, train, and deploy ML models that power search, ranking, recommendations, and retrieval systems. Day to day they work with large datasets, write production-grade Python, run experiments, and collaborate with backend and product teams to ship models that serve real users at scale. Freshers are expected to have strong fundamentals in ML, data structures, and software engineering, plus a willingness to learn from senior engineers.

Role and technical knowledge

1. 1

   Explain the difference between precision and recall. When would you optimise for one over the other in a search ranking system?

   Define both clearly, give a search example where false negatives are costly (missing relevant documents) versus false positives (irrelevant results), and mention F1 or business metrics as trade-off tools.

2. 2

   What is the bias-variance trade-off? How do you detect and address high bias or high variance in a model?

   Describe underfitting and overfitting, learning curves, and remedies like more data, regularisation, or simpler models. Connect to real ML pipelines you have built.

3. 3

   How does a transformer model work at a high level? Why is it useful for search and retrieval tasks?

   Cover self-attention, positional encodings, and how embeddings capture semantic similarity. Mention advantages over RNNs for parallelisation and long-range context.

4. 4

   What is the role of an embedding in a recommendation or search system? How would you evaluate embedding quality?

   Explain that embeddings map items to dense vectors for similarity search. Evaluation can use recall at K, mean reciprocal rank, or downstream task metrics like click-through rate.

Practical scenarios

1. 5

   You are given a dataset of user queries and clicked documents. How would you build a model to rank documents for a new query?

   Outline data cleaning, feature engineering (query-document similarity, click features), training a learning-to-rank model, offline evaluation with NDCG, and online A/B testing.

2. 6

   Your model performs well offline but poorly in production. What could be the reasons and how would you debug it?

   Discuss data drift, training-serving skew, feature leakage, latency issues, and differences in evaluation metrics. Suggest logging, shadow deployment, and monitoring.

3. 7

   How would you handle missing values and categorical features in a dataset for a tree-based model versus a neural network?

   For trees, mention imputation and target encoding; for neural nets, embeddings for categoricals and learned imputation. Highlight cross-validation to avoid leakage.

4. 8

   You need to deploy a model that serves 1000 requests per second with low latency. What steps would you take?

   Cover model optimisation (quantisation, pruning), caching, batching, using a fast inference server, and load testing. Mention trade-offs between accuracy and speed.

Coding and problem solving

1. 9

   Write a function to find the top K most frequent words in a list of strings. Discuss time and space complexity.

   Use a hash map and a heap for O(n log k) time. Explain why sorting all words is less efficient for large n.

2. 10

   Given two sorted arrays, find the median of the combined array in O(log(min(m, n))) time.

   Describe the binary search partition approach. Walk through edge cases like empty arrays and odd/even total length.

3. 11

   Implement a simple gradient descent for linear regression in Python. How do you choose the learning rate?

   Show vectorised code, cost function, and update rule. Explain learning rate tuning via grid search or adaptive methods like Adam.

4. 12

   How would you detect and remove duplicate documents from a large corpus of text?

   Mention hashing (MinHash, SimHash) for near-duplicates, shingling, and Jaccard similarity. Discuss scalability with MapReduce or Spark.

Behavioural and HR

1. 13

   Tell me about a time you worked on a team project where you had to learn a new technology quickly. How did you manage?

   Use STAR format: situation, task, action, result. Highlight self-learning, asking for help, and delivering on time.

2. 14

   Why do you want to work as a Machine Learning Engineer in an AI-first product company?

   Connect personal interest in ML to the company's mission of improving search or knowledge work. Show you understand the role's impact on users.

3. 15

   Describe a situation where your model or code did not work as expected. What did you do?

   Focus on debugging steps, learning from failure, and iterating. Show resilience and a systematic approach.

4. 16

   Where do you see yourself in three years in this role?

   Express growth in technical depth, ownership of ML components, and contribution to production systems. Align with the company's engineering ladder.