Uniphore Machine Learning Engineer Interview: Rounds, Questions and Preparation

What the job involves

A Machine Learning Engineer in AI and deeptech builds, trains, and deploys models that turn speech, text, and other data into usable product features. Day to day you write Python, clean and label data, run experiments, evaluate models, and work with backend engineers to ship models into production. At conversational AI companies you often focus on ASR, NLP, and real-time inference under latency and cost constraints.

Role and technical knowledge

1. 1

   Explain the difference between precision and recall. When would you optimise for one over the other in a speech or text classification product?

   Define both clearly with a confusion matrix example. For ASR keyword spotting or fraud detection, say which error is costlier and how you would tune the threshold.

2. 2

   What is overfitting and how do you detect and reduce it in a deep learning model?

   Mention train vs validation loss curves, regularisation, dropout, data augmentation, early stopping, and more data. Give a concrete example from an NLP or speech task.

3. 3

   How does a transformer differ from an RNN for sequence tasks like speech recognition or intent classification?

   Cover parallel processing, attention, long-range dependencies, and training speed. Note that RNNs handle streaming well while transformers need chunking or specialised attention for real time.

4. 4

   What metrics would you use to evaluate a conversational AI model for intent detection and entity extraction?

   Mention accuracy, macro F1, per-class precision and recall, and exact match for entities. Discuss class imbalance and why overall accuracy can mislead.

Practical scenarios

1. 5

   You are given a dataset of customer support calls with noisy transcripts. How would you prepare the data for training an intent classifier?

   Talk about cleaning, normalisation, handling missing or mislabelled data, class balance, train/validation/test splits, and possibly weak supervision or active learning.

2. 6

   Your model performs well offline but is slow in production. How would you reduce inference latency?

   Suggest model compression, quantisation, pruning, distillation, batching, caching, and choosing a smaller model. Mention measuring latency and cost per request.

3. 7

   A model you deployed starts giving worse predictions after a few weeks. What do you check first?

   Discuss data drift, concept drift, logging and monitoring, retraining triggers, and comparing live metrics to validation metrics. Mention feedback loops and label delay.

4. 8

   How would you build a simple pipeline that takes raw text, trains a classifier, and saves the model for serving?

   Describe steps: ingest, preprocess, vectorise, train, evaluate, save with joblib or torch.save, and expose via a Flask or FastAPI endpoint. Mention Docker and versioning.

Behavioural and teamwork

1. 9

   Tell me about a project where you had to learn a new ML tool or technique quickly. How did you approach it?

   Use STAR structure. Focus on the learning method, resources used, and the measurable outcome. Show curiosity and self-direction.

2. 10

   Describe a time you disagreed with a teammate on a technical approach. What happened?

   Show that you listen, use data or experiments to decide, and keep the team goal first. Avoid blaming anyone.

3. 11

   How do you explain a complex ML result to a non-technical stakeholder?

   Give an example where you avoided jargon, used a simple analogy or visual, and tied the result to a business or user outcome.

4. 12

   Why do you want to work in AI and deeptech, and what do you know about our products?

   Connect your interest to real use cases like conversational AI, speech analytics, or automation. Mention specific product areas you have read about, without claiming inside knowledge.

HR and fresher-specific

1. 13

   Walk me through your resume and highlight the ML work you are most proud of.

   Keep it to two minutes. Focus on problem, your role, tools, results, and what you learned. Be ready for follow-up questions on any line.

2. 14

   This role may involve working on data privacy sensitive speech and text data. How would you handle that responsibly?

   Mention anonymisation, access control, consent, bias checks, and following company and legal guidelines. Show awareness of fairness and user trust.

3. 15

   Where do you see yourself in two years, and how does this role fit?

   Show realistic growth toward owning models end to end, deeper specialisation in NLP or speech, and contributing to production systems. Avoid saying you want to leave for a different field.

4. 16

   Do you have any questions for us?

   Ask about team structure, model deployment stack, mentorship for freshers, or how success is measured in the first six months. Avoid questions only about salary at this stage.