ML Engineering interview question · The ML lifecycle · mid level
A model scores 95% accuracy offline but users say it is useless. What do you check first?
Free to read in full. This is 1 of the 150 questions in the ML Engineering and MLOps interview pack, with 193 diagrams. Get the pack, ₹999 · $20
Why it is asked
A classic debugging question: the interviewer wants a calm, ordered checklist starting with the most likely cause.
The answer
First check what 95% is being compared with: if only 5% of cases are positive, a model that always says "no" also scores 95% accuracy while catching nothing. Accuracy is the share of all predictions that are right. When one class is rare, accuracy hides failure. With 50 positives in 1,000 cases, the always-no model scores 0.95 accuracy, 0.0 recall (it finds none of the positives) and 0.5 balanced accuracy (the average recall over both classes, where 0.5 is coin-flip level). My checks, in order: 1. Class balance and the right metric. Compare with the always-majority baseline. Use precision (of the cases we flag, how many are real), recall (of the real cases, how many we flag), or precision in the top k if users only see a short list. 2. Leakage. Does a feature contain the answer, such as a "refund_issued" field when predicting fraud? Are rows from the same customer in both train and test? Leakage gives high offline scores that vanish live. 3. Training-serving skew. Is a feature computed differently in production, missing, or arriving late? Compare the live input distribution with the training one. 4. Offline data versus real use. Was the test set old, from one region, or cleaner than what users send? 5. The user's real task. "Useless" may mean the errors land on the cases users care about most, such as large orders. Read 50 real failures and slice the errors by segment. Only after these checks would I change the model itself.
What the interviewer listens for
- Accuracy on an imbalanced label can equal the do-nothing baseline.
- Recall, precision and balanced accuracy expose it.
- Leakage and training-serving skew are the next suspects.
- Check that the test set looks like real usage.
- Read real failures before touching the model.
Follow-ups the interviewer may ask
- Which metric would you report for a 1% positive rate? Hint: precision and recall at the operating threshold, or average precision (area under the precision-recall curve).
- How do you detect leakage quickly? Hint: look for a single feature with near-perfect importance, and check feature timestamps against the prediction time.
- How do you compare live inputs with training inputs? Hint: per-feature statistics or PSI (population stability index, a score of how far a distribution has moved) between the training sample and a recent serving sample.
Common mistakes
- Trusting accuracy without checking the class balance.
- Retraining a bigger model before ruling out data problems.
- Asking users for vague feedback instead of reading specific failures.
Written by Roni Das, who builds distributed systems and AI systems for a living.
Learn it properly
This is taught step by step in the course lesson The Score That Lied: A Random Split Against a Split in Time.
One question is not enough for the interview.
This is 1 of the 150 questions in the ML Engineering and MLOps pack
Each question has why it is asked, a full answer, what the interviewer listens for, the follow-ups, the common mistakes and a diagram (193 diagrams in all). A PDF you keep, junior to staff level.
India: UPI, cards or netbanking through Razorpay. Elsewhere: PayPal. Download at once from My packs.
More free questions from this pack
- How do you train on a dataset that does not fit in memory? (junior)
- A long distributed training job dies at hour 20. How do you make training fault tolerant? (senior)
- How does request batching on a GPU trade latency for throughput? (mid)
- An online model metric dropped 8% overnight. How do you debug it? (staff)
Free questions from the other packs
- What do temperature and top-p control, and what values would you use for data extraction versus brainstorming? (AI Engineering)
- What is hybrid search, and how do you merge keyword and vector results with Reciprocal Rank Fusion? (AI Engineering)
- Users ask "list all X" questions and RAG returns only five items. Why, and what do you do? (AI Engineering)
- How do you check that an LLM judge agrees with human reviewers? (AI Engineering)
- Cut the cost of an LLM feature serving 10 million requests a day by 60% without hurting quality. Walk through your plan. (AI Engineering)
- Why are LLMs bad at counting letters or reversing a word? (GenAI and LLM)
- Write out scaled dot-product attention with the shape of every tensor. (GenAI and LLM)
- How does speculative sampling keep exactly the target model's distribution? (GenAI and LLM)
- What is GRPO, and how do verifiable rewards train reasoning models? (GenAI and LLM)
- Design a system that answers questions from 300-page contracts accurately. (GenAI and LLM)
- Describe the basic agent loop. (Agentic AI)
- What are pass@k and pass^k, and why do they matter for agents? (Agentic AI)
- How does prompt injection become dangerous once the agent has tools? (Agentic AI)
- Design the security architecture for an agent that can read and send company email. (Agentic AI)
- Design a customer support agent for an e-commerce company. (Agentic AI)
- What happens, step by step, when an LLM server handles one request? (Inference Engineering)
- Derive the KV cache size formula and work it out for Llama 3 70B. (Inference Engineering)
- What is PagedAttention? (Inference Engineering)
- When does speculative decoding stop helping or make things worse? (Inference Engineering)
- Design a service to serve Llama 3 70B at 10 QPS with latency targets. (Inference Engineering)
- You need to show delivery times for 20 cities. Would you use histograms, box plots, violin plots or ECDFs, and why? (Data Science)
- A test is 99% accurate and you test positive. What is the chance you have the disease? (Data Science)
- What share of first-time buyers make a second purchase within 30 days? Write the SQL. (Data Science)
- How would you forecast daily orders for the next four weeks, and how do you know the forecast is any good? (Data Science)
- A coupon A/B test shows a small average lift. How do you find which users the coupon actually works for, and who should get it next time? (Data Science)
- Compute the cosine similarity of a = (3, 4) and b = (4, 3) by hand. What does the result mean? (Maths for ML)
- A batch of 32 MNIST images goes into a dense layer with 128 units. What are the shapes, and what is the rule for when two matrices can be multiplied? (Maths for ML)
- A disease affects 1% of people. A test catches 90% of sick people (sensitivity) and wrongly flags 9% of healthy people (false positive rate). You test positive. What is the chance you are sick? (Maths for ML)
- Why do we not use mean squared error for classification with a sigmoid output? Compare the gradients numerically. (Maths for ML)
- Compute scaled dot-product attention by hand for two tokens with d_k = 2. (Maths for ML)
- A customer says: "We want AI for our support team." What do you do in the first meeting? (Forward Deployed Engineer)
- The customer's admin offers you a global admin account "to save time". What do you ask for instead, and why? (Forward Deployed Engineer)
- Your connector writes tickets into the customer's system. A request timed out, the connector retried, and now there are duplicate tickets. How do you make the writes safe to retry? (Forward Deployed Engineer)
- The customer wants RAG over 10 million documents, all behind their firewall. Size it for me on the whiteboard. (Forward Deployed Engineer)
- One deployment of your AI platform must serve 50 business units of the same enterprise. How do you design multi-tenancy? (Forward Deployed Engineer)