Interview questions for a tech lead.

The questions.

1. 01

   technical medium · distributed-systems

   Describe an event-driven architecture you've designed or significantly contributed to. Delivery guarantees? Failure modes you had to plan for?

   Strong answer

   Names specific system + team + problem. Discusses trade-offs (throughput vs ordering, latency vs durability) explicitly. Cites a real failure mode they lived through. Seniors: how they'd evolve the design today.

   Weak answer

   Describes a textbook Kafka setup without team or context. Vague on delivery guarantees. No real failure mode.
2. 02

   technical baseline · concurrency

   How do you handle concurrent writes to the same resource (e.g., two users editing the same document)?

   Strong answer

   Names a specific strategy — optimistic locking, CRDTs, operational transform, or last-write-wins with explicit trade-off discussion. Mentions how they'd expose conflicts to the user.

   Weak answer

   Hand-waves about 'locks.' Doesn't distinguish optimistic vs pessimistic. No user-facing conflict resolution.
3. 03

   situational hard · dates-vs-quality

   Your CEO commits you to a public launch date your team can't safely hit. Walk me through what you say to them.

   Strong answer

   Separates date concern from reputation concern. Proposes options (scope cut, phased rollout, extended beta). Doesn't just say no. Seniors: surfaces trade-offs so the CEO chooses.

   Weak answer

   Either agrees and burns out the team, or refuses outright without offering alternatives.
4. 04

   communication baseline · explaining

   Explain why ACID transactions matter to a non-technical co-founder asking 'can't we just use a NoSQL database?'

   Strong answer

   Concrete examples (double-charging, lost orders). No jargon drop. Acknowledges valid NoSQL use cases. Seniors: helps them think, doesn't just explain.

   Weak answer

   Jumps to CAP theorem. Condescending tone. Doesn't engage with the question as asked.
5. 05

   situational medium · teammate-conflict

   A teammate ships a change you think is wrong after the fact. Already in production. What do you say, what do you do?

   Strong answer

   Separates the product concern from the relationship concern. 1:1 first. Focuses on systemic change (PR review, tests) — not on blame.

   Weak answer

   Avoids the conversation or jumps to escalation. Assigns blame. No systemic improvement suggested.
6. 06

   technical medium · api-design

   You're designing an API endpoint that returns a user's feed. Pagination strategy? Why?

   Strong answer

   Discusses cursor vs offset-based pagination · knows when each is appropriate · mentions stability under inserts · sets sensible default + max page size · considers ETag/If-None-Match for caching.

   Weak answer

   Picks offset pagination by default without considering stability. No max page size. No caching thought.
7. 07

   culture medium · candor

   Your manager proposes a technical direction you think is wrong. Three other senior engineers nod along. What do you say in the meeting, and what do you do in the next 48 hours?

   Strong answer

   Speaks up in the meeting (even tentatively). Grounds disagreement in reasoning, not ego. Follows up in writing. Separates 'disagree' from 'won't execute.'

   Weak answer

   Stays silent in meeting, vents to a peer later. Or raises it aggressively without offering an alternative.
8. 08

   technical baseline · postgres

   You have a Postgres table with 500M rows of loan applications. A query filtering by applicant_id (indexed) and status (not indexed) takes 3 seconds. Walk me through how you'd diagnose and fix this.

   Strong answer

   Asks about workload before jumping to solutions · proposes EXPLAIN ANALYZE · mentions partial or composite indexes, not just 'add an index' · seniors reference statistics, vacuum, or plan cache unprompted.

   Weak answer

   Immediately says 'add an index on status.' Doesn't ask about cardinality. Doesn't mention the query planner.
9. 09

   technical hard · scaling

   Walk me through how you'd scale a write-heavy service from 100 to 10,000 writes/sec. What breaks first?

   Strong answer

   Starts with profiling current bottleneck, not with 'add Kafka.' Acknowledges write-amplification, index bloat, replication lag. Discusses shard keys + hot-partition mitigation.

   Weak answer

   Jumps to 'we'd add a queue and scale horizontally' without identifying where the current bottleneck is.
10. 10

    technical hard · observability

    What does a useful logging + metrics + tracing stack look like for a service with 100M requests/day?

    Strong answer

    Distinguishes logs/metrics/traces by use case. Knows sampling strategies. Structured logs. Cardinality discipline. Mentions specific tools without being dogmatic about them.

    Weak answer

    Conflates logs and metrics. No sampling strategy. Suggests logging every request body (cardinality disaster).