10 Plaid Software Engineer (New Grad) Interview Questions (2026)

Outline

Plaid wants product thinkers. Pick a specific consumer fintech app you have used (or a developer-facing API you have integrated) and articulate why the underlying connectivity layer matters. Avoid generic 'fintech is interesting.' Show you understand that Plaid sits between banks and apps and that reliability and consent are the product.

Outline

STAR. Plaid ships to regulated financial customers — they want engineers who can spot when 'fast and broken' is unacceptable, but who can also pragmatically ship the 90% solution. Show your reasoning: what was at stake, what evidence informed the call, who you talked to, what guardrails you added. Avoid stories where the tradeoff was reckless.

Outline

STAR. Plaid is an infrastructure company; reliability stories land well. Concrete metrics (error rate before/after, time-to-recover, SLO). Show ownership — you noticed, you fixed, you documented, you set up monitoring so it would not recur. Avoid hero stories where you were the only one who could fix it.

Outline

Kadane's algorithm wrapped in a domain story. Track currentSum (reset to current if previous went negative) and maxSum. O(n) time, O(1) space. Discuss edge cases: all negatives (a single transaction is the answer), empty list. Follow-up: return the actual transaction range, or handle multiple accounts.

Outline

Min-heap of size K storing (value, list-index, value-index). Pop, push the next element from the same list. O(N log K) where N is total elements. Alternative: divide-and-conquer pairwise merge (same complexity, sometimes easier to write). Walk through one merge step with a small example.

Outline

DFS with a running sum, track max at leaf. O(n) time, O(h) space for recursion. Variant: maximum path sum between any two nodes (harder — recurse returns max gain through current node, update global max). Pick the variant they actually asked, then walk through the recursion carefully.

Outline

Hash map of count + min-heap of size K keyed by count. On each event: increment map, push to heap, pop if size > K. O(log K) per event. For very high cardinality, mention count-min sketch as a memory tradeoff. Plaid likes this because it maps to anti-fraud signal aggregation.

Outline

Layered approach. First: how do customers measure missing — gaps in their DB vs delayed retries? Then check our send side (DB log, queue depth, retry config). Then network (DNS, TLS, customer endpoint reachability). Then customer side (their endpoint dropping, returning 5xx silently). Mention idempotency tokens + signed-delivery receipts as long-term fixes. Plaid asks this because webhooks ARE the product.

Outline

Single-pass string builder. Skip whitespace and dashes, validate remaining characters are digits. Clarify: should we validate length per country? Mod-10 checksum? O(n) time, O(n) space for the new string. Edge cases: empty after stripping, mixed alphanumeric, unicode digits. Plaid loves input-normalization questions because banks send messy data.

Outline

Sliding window log, fixed window counter, or token bucket. Sliding window log: store timestamps per key, evict expired on each request. Token bucket: refill rate + capacity, decrement per request. Discuss tradeoffs (memory vs precision). Mention distributed concerns (Redis with TTL) only if asked. Plaid asks this because banks rate-limit their integrations.