Writing
Writing across distributed systems, streaming infrastructure, cloud-native platforms, and AI-native systems — published on serverless.fyi and snackonai.com.
Distributed Systems
A deep dive into Jepsen's role in testing distributed systems — how to surface consistency bugs that only appear under real network failure conditions.
Dec 2024 · serverless.fyi →Keeping all systems in sync — how vector clocks solve causality tracking in distributed systems where wall clocks cannot be trusted.
Dec 2024 · serverless.fyi →How distributed file systems coordinate storage across many nodes — the design tradeoffs, consistency models, and real-world implementations.
Dec 2024 · serverless.fyi →Streaming
Reflections from Austin, Texas — how Apache Kafka and Apache Flink are converging with AI to define event-driven architectures for the next decade.
Sep 2024 · serverless.fyi →Harnessing the power of multi-tenancy in Apache Pulsar for secure and scalable data streaming — namespaces, policies, and isolation in practice.
Nov 2024 · serverless.fyi →Demystifying the streaming protocol stack — what RTMP, HLS, and RTSP each do, when to use them, and how they fit into modern media infrastructure.
Dec 2024 · serverless.fyi →Cloud Infrastructure
Optimizing script usage in infrastructure as code — a practical guide to choosing the right Terraform escape hatch for your use case.
Dec 2024 · serverless.fyi →How next-generation AI is learning to comprehend, query, and reason over structured relational data — and what it means for the database layer.
Nov 2025 · serverless.fyi →The tiny packet that runs the internet — a deep look at DNS header structure, flag fields, and how every network request begins.
Oct 2025 · serverless.fyi →AI + Infrastructure
Feature stores are often oversold. Here's what they actually do, where they fail, and how to think about the real-time feature engineering layer in production ML systems.
snackonai.com →The KV cache is the most underappreciated layer in LLM inference. LMCache and SGLang show what a proper caching infrastructure for large models actually looks like.
snackonai.com →Retrieval-Augmented Generation isn't a pattern — it's the foundational data access layer for AI systems. Here's how to think about building it right.
snackonai.com →What does it actually take to coordinate multiple AI agents in production? Gas Town explores the orchestration layer — routing, state, and reliability at agent scale.
snackonai.com →The simplest agentic architecture is often the best one. Why a tight feedback loop beats elaborate orchestration frameworks for most real-world agent tasks.
snackonai.com →TensorRT-LLM is widely misunderstood. It's not just a quantization tool — it's a full inference compiler that fundamentally changes how LLMs run on GPU infrastructure.
snackonai.com →
"The hardest part of systems design isn't choosing the right architecture.
It's knowing which tradeoffs you're actually making."
— Mohinish