10 IBM Research Software Engineer (New Grad) Interview Questions (2026)

Outline

STAR: pick a real example — presenting research to undergrads, explaining a system to a PM or business stakeholder, writing a blog post. Show you can compress complex ideas without losing accuracy. Critical at IBM Research where you bridge labs and product.

Outline

Tie to a specific research thrust — quantum, AI safety, hybrid cloud, accelerated computing, materials simulation. Show you understand the IBM Research model: long-horizon problems, papers + product, bridging fundamental research to enterprise customers. Avoid 'I want to work on cool things' alone.

Outline

STAR: pick a real story where data or feedback changed your belief about a design or hypothesis. Show intellectual honesty and openness to being wrong. Research labs value scientific thinking — confirmation-bias-resistant candidates stand out.

Outline

Two standard approaches: (1) Kahn's algorithm — compute in-degrees, start from zero-in-degree nodes in a queue, decrement neighbors' in-degrees as you process. (2) DFS with post-order — reverse the post-order. Detect cycles in either. O(V+E) time, O(V) space.

Outline

Factorial number system. The kth permutation can be constructed digit-by-digit: at each step, divide remaining k by (remaining n - 1)! to choose the index of the next element. Decrement k by the quotient * factorial. O(n^2) for the index-removal, O(n log n) with a more advanced data structure. Walk through n=3, k=4 explicitly.

Outline

Hash map of word → index. For each word, split it at every position; if the first half is a palindrome and the reverse of the second half is in the map, that's a pair. Symmetric check for the other direction. O(n * k^2) where k is the max word length. Discuss the trie alternative.

Outline

If the matrix is row-sorted and column-sorted, use binary search on the value range. count(target) = number of cells <= target, walk diagonally O(n). Binary search the value range using this count. O(n log(max - min)) time, O(1) space. Alternative: min-heap of size k, O((k + n) log n).

Outline

Pick a project with depth — a thesis, a serious internship, or a substantial open-source contribution. Show the problem, why it was hard, what you tried that failed, what worked, and how you measured success. IBM Research interviewers will probe deeply — be ready for follow-ups on every decision.

Outline

Pick a paper actually relevant to the team you are interviewing for. Summarize crisply (problem, approach, results, limitation). End with a concrete extension you would try. Show you read actively, not passively. Interviewers will follow up on the limitation you identified.

Outline

Read the paper carefully (especially methods and supplementary). Check for code/data release. Build a minimal version of the experiment. Verify hyperparameters and seeds. Compare your results to the paper's. Discuss why minor discrepancies happen (versioning, hardware, randomness) and when they signal a real issue. Critical research-skill question.