ML Coding Interview Questions (2026)

Hands-on ML coding rounds: implement algorithms from scratch in NumPy, debug training pipelines, and write production-quality ML code under time pressure. This bank has 83 ml coding questions (24 with free model answers).

Other topics: ML Theory · ML System Design · Behavioral · All questions

Practice this question with an AI interviewer

Get asked follow-ups live, then receive a scored report — like a real MLE interview loop.

Start AI mock interview

#35 Given Item Counts, How Would You Return the K Most Frequent Values? ★
#36 How Would You Build an O(1) Least-Recently-Used Cache? ★
#37 How Would You Code Stable Softmax, Cross-Entropy, and Their Gradient in NumPy? ★
#38 How Would You Find the First Sorted Position Meeting a Lower-Bound Condition? ★
#41 How Would You Train Logistic Regression Using Gradient Descent from Scratch? ★
#294 Implement a BPE tokenizer from scratch: train merge rules, then encode and decode without external libraries. ★
#295 Implement scaled dot-product self-attention in NumPy or PyTorch with an optional causal mask. ★
#296 Implement a correct multi-head self-attention forward pass with projections, reshaping, and masking. ★
#297 Build a full transformer block from scratch: multi-head attention, feed-forward network, residuals, and layer norm. ★
#298 Implement FlashAttention-style tiled attention without ever materializing the full N×N matrix. ★
#299 Hand-write a KV cache and single-step decode. ★
#300 Implement top-k, top-p, and temperature sampling over a logits vector. ★
#301 Implement top-k MoE routing with a shared expert in PyTorch and point out the efficiency and correctness traps.
#302 Implement a top-k MoE router, then contrast auxiliary-loss load balancing with a loss-free approach.
#303 Hand-write a miniGPT.
#304 Debug a Transformer: find the classic bugs. ★
#305 Implement k-means clustering on a shared doc. ★
#306 Implement logistic regression from scratch in Python with no ML libraries.
#307 In NumPy, compute minADE and minFDE for multi-modal trajectory predictions with variable-length ground truth — no Python loops.
#308 Implement non-maximum suppression, then vectorize it.
#309 Design and implement an in-memory key-value store supporting set plus transactional begin, commit, and abort. ★
#310 Build an in-memory key-value store with SET/GET/DELETE, then add transactions with BEGIN/COMMIT/ROLLBACK including nested transactions. ★
#311 Implement the block manager for a paged KV cache: allocate, append, free, and copy-on-write prefix sharing.
#312 Implement an LRU cache with O(1) get/put, then add TTL. ★
#313 Design an LRU cache from scratch and explain how multiple users can use it concurrently with consistency.
#314 Implement a token-bucket rate limiter, then make it distributed. ★
#315 Build a rate limiter whose requirements grow every ten minutes (per-tenant limits, burst allowances, sliding window) without the code collapsing.
#316 Find every concurrency bug in a BankAccount class whose deposit does read → sleep(0.1) → write, then provide a thread-safe version via subclass/wrapper locking.
#317 On CoderPad, run unit tests against unfamiliar Python code and debug why they fail.
#318 Hand-write a parallelized sort.
#319 Design a restorable (checkpointable) iterator.
#320 Review code: find the issues in an input-field implementation.
#321 Write the Python to run an LLM call over 50,000 documents given ~100 concurrency and occasional 429s and timeouts.
#322 Write an async batch processor over an API with concurrency limits, retries with jitter, and error isolation. ★
#323 Build an in-memory database: SET/GET/DELETE first, then filtered scans, TTL with timestamps, then file compaction.
#324 Implement a streaming top-k with a bounded-memory sketch, and a sliding-window rate counter.
#325 Find the top-K most frequent items in a high-volume event stream with bounded memory.
#326 Return the k most frequent event IDs from an unbounded stream at any point, with bounded memory. ★
#327 From a stream of user actions, return the k most engaged-with items — then explain why a heap solution could be wrong in production.
#328 Implement the top-k most frequent search queries over a large query log, then explain what breaks when the log becomes an unbounded stream across many machines.
#329 Implement a minimal agent loop with tool dispatch, error handling, and a step budget.
#330 Build the core of an editor text buffer with efficient arbitrary insert/delete and fast line lookup.
#331 Given a repository snapshot (path → content), build a Merkle tree and return which files changed between two snapshots without comparing every file's content.
#332 Implement an efficient ring buffer for high-rate sensor data with a fixed memory budget.
#333 Write a SQL query to compute 7-day rolling conversion by acquisition channel.
#334 Implement a thread-safe batching logger: many producers call log(msg); a background thread flushes batches of up to 100 messages every second or when full.
#335 Diagnose a Spark job joining a 2TB fact table to a 50GB dimension table where one straggler task runs 100x longer than the rest.
#336 Build the task-lifecycle core of an annotation platform, then add k-annotator consensus, priority re-review, and annotator cooldowns.
#337 Given annotation sessions as (start, end) timestamps, return the peak number of concurrent annotators and the intervals at peak load.
#338 Simulate plant infection / cellular automaton on a grid: compute days to stabilize, an immune-cell variant, and an immunity countdown (5 progressive parts).
#339 Implement GPU credit operations (create_grant / subtract / get_balance) with expiry priority and out-of-order operations, then optimize for production.
#340 Build a text editor in 4 parts: buffer ops, dual-stack undo/redo, AI autocomplete with Trie + word frequencies, and real-time collaboration/OT.
#341 Schedule data-labeling tasks across (task, model) and (task, human) pairs.
#342 Add two floating-point numbers given as strings and return the sum as a string, without eval or float conversion.
#343 Determine whether a linked list is a palindrome in O(1) space.
#344 Tag substrings with XML italic tags and remove nested tag pairs inside the longest pair (extended merge intervals); write edge-case tests.
#345 Find the maximum contiguous free slot for two people, then the max slot if you may flip one busy slot to free.
#346 Solve the water jug problem with BFS or DFS.
#347 Implement weighted random sampling with follow-ups, and write a test function proving the non-deterministic solution works.
#348 Serialize and deserialize a binary tree. ★
#349 Merge two sorted lists. ★
#350 Reverse a linked list. ★
#351 Compute the edit distance between two strings. ★
#352 Debug a program that models infection spread across a social graph.
#353 Refactor ~120 lines of working but messy code with passing tests, improving the architecture without breaking them.
#354 A model runs fine in FP32 but produces garbage after conversion to FP16 — debug it.
#355 Debug a CUDA kernel running 10x slower than expected, without running it: name the usual suspects and how to confirm each.
#356 Find and fix a double-counting bug in a function that tallies values in a HashMap.
#437 Implement a random forest classifier from scratch without scikit-learn, using bagged trees and majority vote
#438 Implement TF-IDF vectorization from scratch with NumPy
#439 Implement beam search decoding over next-token logits from scratch
#440 Implement the batch normalization forward and backward pass in NumPy
#441 Implement precision, recall, and F1 from raw predicted labels in NumPy, without sklearn
#442 Build a k-nearest neighbors classifier from scratch with Euclidean distance and a tie-breaking rule
#443 Implement multivariate linear regression via the normal equation in NumPy and handle singular matrices
#444 Implement a single neuron's forward and backward pass with sigmoid activation and MSE loss in NumPy
#445 Implement PCA via SVD in NumPy without using sklearn
#446 Implement reservoir sampling to sample k items from a data stream in Python
#447 Given a user event log DataFrame, compute per-user rolling 7-day conversion features in pandas
#448 Implement 2D convolution with configurable stride and padding in NumPy
#449 Implement Gaussian Naive Bayes from scratch in NumPy
#450 Implement the Adam optimizer update step from scratch in NumPy
#451 Debug a U-Net segmentation model live in PyTorch/NumPy and find the bugs