A curated reading list of technical engineering blog posts from leading technology companies, organized by topic. Each entry includes a short summary to help you decide what to read next.
All posts link to the original source. Nothing here is reproduced or owned by this repository.
- Artificial Intelligence and Machine Learning
- Distributed Systems and Infrastructure
- Security and Privacy
- Game Engineering and Real-Time Systems
- Developer Experience and Tooling
- Data Science and Experimentation
| Company | Engineering Blog |
|---|---|
| Meta | engineering.fb.com |
| Spotify | engineering.atspotify.com |
| Riot Games | engineering.riotgames.com |
| Go / Google | go.dev/blog |
| Uber | eng.uber.com |
| # | Title | Company | Description |
|---|---|---|---|
| 01 | An Organizational Second Brain: Building an AI That Learns From Experts | Meta | How Meta built a system to capture and scale domain expertise from human experts into a queryable AI model. |
| 02 | When Can LLMs Replace Humans in A/B Tests? | Spotify | An honest look at where LLM-based predictions hold up as substitutes for human behavioral data in controlled experiments. |
| 03 | Better Experiments with LLM Evals — A Funnel, Not a Fork | Spotify | How Spotify structures LLM evaluation pipelines to surface quality issues early without slowing down iteration. |
| 04 | Our Multi-Agent Architecture for Smarter Advertising | Spotify | The design behind a multi-agent system that coordinates specialized AI agents to handle advertising decisions end to end. |
| 05 | From User Sequences to Scaling Laws: Meta's Ads Ranking | Meta | How Meta built a multi-stage ranking architecture for ads that applies scaling law principles to recommendation systems. |
| # | Title | Company | Description |
|---|---|---|---|
| 06 | ZGateway: Learnings from Putting a Proxy in Front of ZippyDB | Meta | What Meta learned adding a proxy layer to their distributed key-value store, including the tradeoffs around latency, consistency, and operational complexity. |
| 07 | MetaRoCE: A New RDMA Transport Built for AI-Scale Ethernet | Meta | Meta's custom RDMA networking protocol designed to meet the demands of large-scale AI training clusters over standard Ethernet. |
| 08 | MTIA 300: Meta's First Training Chip with Built-in NICs | Meta | A technical overview of Meta's second-generation AI training accelerator, the first to integrate network interface cards directly into the chip. |
| 09 | Indexing the Data Lake for Online Point Queries | Spotify | How Spotify built an indexing layer on top of their data lake to serve low-latency point queries without duplicating data into a separate store. |
| 10 | How Uber Manages Millions of Consistent Kafka Messages | Uber | Uber's approach to guaranteeing exactly-once message delivery and ordering across high-throughput Kafka pipelines. |
| # | Title | Company | Description |
|---|---|---|---|
| 11 | How We're Building Scam Alert on WhatsApp With End-to-End Encryption | Meta | The cryptographic design behind WhatsApp's scam detection feature that works without breaking end-to-end encryption or exposing message contents to Meta. |
| 12 | Riot's Vanguard: Kernel-Level Anti-Cheat Design Decisions | Riot Games | Why Riot Games chose to run their anti-cheat system at the kernel level, the security implications of that decision, and how they defend against the risks it introduces. |
| # | Title | Company | Description |
|---|---|---|---|
| 13 | Fixing the Internet for Real-Time Applications | Riot Games | How Riot built their own global network backbone to reduce latency and packet loss for League of Legends players, rather than relying on the public internet. |
| 14 | League of Legends' Architecture — A Deep Dive | Riot Games | A detailed look at the server-side architecture behind one of the world's most-played games, covering matchmaking, game servers, and real-time state synchronization. |
| # | Title | Company | Description |
|---|---|---|---|
| 15 | Portal by Spotify Cut My Claude Code Token Usage by 90% | Spotify | How Spotify's internal developer tool Portal dramatically reduced AI coding agent token consumption by optimizing the context sent to the model. |
See CONTRIBUTING.md for how to suggest or add a blog entry.
The summaries and structure in this repository are released under the MIT License.
All linked articles remain the property of their respective publishers and authors.