HackTheRounds Interview Experiences

OpenAI Software Engineer Interview Experience (2025) - Rejected

Full interview breakdown for OpenAI SWE role: recruiter screen, technical phone screen with coding questions, and 4 round virtual onsite including system design

By Anonymous ยท 2026-01-22

Background

Applied for Software Engineer role at OpenAI. Had 4 YOE, previously at a large tech company working on backend infrastructure. Applied via referral which I think helped get the initial response faster.

Technical Phone Screen (60 min)

Problem ( 40 min): Given a key-value store, implement serialization and deserialization that handles multiple value types. The followup was to handle encoding and decoding of nested structures.

[[problem/345?company=8|KV Store Serialization]]

The interviewer was friendly and gave hints when I was going down the wrong path. We had 10 min at the end for questions.

Virtual Onsite (5 rounds)

Round 1: Coding

Graph problem involving task scheduling with dependencies. Basically needed topological sort but with some twists around handling priorities and cycles. The interviewer cared a lot about edge cases and error handling.

Round 2: System Design

Asked to Design Slack from end-to-end.

[[problem/352?company=8|Design Slack]]

I started by summarizing the overall components: - User, Messages, Channels, Notifications, and Search

After fulfilling the basic requirements and adding the schemas, I was asked to address how we can scale: - Fan-out, caching, database sharding

The interviewer specifically asked, "How would you handle fan-out for large channels?"

I spoke about read-time fan-out and added the performance trade-off analysis:

If we choose write-time fan-out, we get better latency for readers but worse scalability for large channels, so I would rather push fan-out to read time.

The interviewer then asked about adding real-time delivery, so I added a notification queue and asynchronous push mechanism, basically event-based messaging system.

They go deep on backend domain specifics, so would recommend going through Alex Xu's system design book at the very least.

Round 3: Coding

Implement a social network data structure with snapshot functionality. Need to handle: - Adding/removing friends - Taking snapshots of the current state - Querying friend lists at any snapshot

[[problem/343?company=8|Social Network with Snapshots]]

Tested understanding of data structure design and how to handle versioning. Finished with time to spare and added some test cases.

Round 4: Behavioral

  • Time you disagreed with a technical decision and how you handled it
  • How do you approach ambiguous problems?
  • Tell me about a project that failed
  • Why AI/ML? What concerns do you have about AI safety?

Make sure you develop a strong AI safety opinion.

Round 5: Experience Deep Dive

Did a deep technical dive into a large-scale backend offline data pipeline I built at my current company: - Architecture decisions and why - What I would do differently - How we handled incidents - How I worked with other teams

Very conversational, felt more like a peer discussion than an interview. They will typically probe how you worked with other teams the most and some of the challenges in cross-functional alignment.

Result

Got rejected a week later, but it was a great experience. I will try again next year.