Openai·Software Engineer·Onsite - System Design / Architecture·Staff

StaffPrefer not to say

Jun 2026Remote

Summary

System design round at OpenAI covering a full-scale activity feed from scratch. The scope was massive and I think I only got through maybe 60% of what they wanted before time ran out.

Questions Asked (1)

Design a large-scale activity and news feed system similar to Facebook or Instagram, covering data modeling, feed generation, ranking, storage, pagination, deduplication, ads, multi-device sync, and scaling to billions of users.

System DesignData ModelingTechnical Trade-offs

Author's notes

This question is basically a full system design career arc compressed into 45 minutes.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying requirements and scale, then design a high-level architecture that separates feed generation (write path) from feed consumption (read path). Dive into data modeling, storage choices, and ranking algorithms, and discuss trade-offs for pagination, deduplication, ads, and multi-device sync. Finally, address scaling strategies like sharding, caching, and CDNs.

Pro tip: Emphasize the push-pull hybrid model for feed generation: push (fan-out on write) for active users and pull (fan-out on read) for celebrities, as it balances latency and cost. Also, mention that ranking should be a separate service that can be iterated independently.

1. Clarify Requirements and Scale

Ask questions to understand functional and non-functional requirements: number of users, read/write ratio, latency SLAs, consistency needs, and features like ads and multi-device sync.

2. High-Level Architecture

Outline the main components: feed generation service, feed storage, ranking service, and APIs for clients. Discuss push vs. pull vs. hybrid approaches for feed generation.

3. Data Modeling and Storage

Design schemas for activities, news feed entries, and user graphs. Choose storage solutions (e.g., SQL vs. NoSQL, graph DB) and discuss sharding and replication.

4. Feed Generation and Ranking

Explain how to generate and rank feeds: use a combination of precomputed feeds and on-the-fly ranking. Discuss ranking signals (recency, affinity, engagement) and ML models.

5. Scaling and Optimizations

Address scaling to billions: caching, CDNs, pagination strategies (cursor-based), deduplication, ads insertion, and multi-device sync. Discuss trade-offs and failure handling.

Key Points to Mention

Push-pull hybrid model for feed generation to handle both normal users and celebrities efficiently.
Cursor-based pagination for consistent and scalable feed reading.
Deduplication strategies to avoid showing the same content multiple times.
Ranking service with machine learning models that can be updated independently.
Multi-device sync using a combination of push notifications and delta updates.
Ads insertion as a separate service that blends into the feed without disrupting ranking.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.