Distributed systems, stream processing and AI infrastructure — not as slides and diagrams, but as systems you build from scratch, break, and reason about yourself.
Every guide starts with the concepts and trade-offs, and how systems like Kafka and Flink handle them.
Implement the system in Go, chapter by chapter, from an empty folder to a running cluster.
Kill nodes, drop messages and inject faults, then verify the system still behaves correctly.
Consensus, replication, leases and recovery, built into systems that survive failures.
Inspired by Kafka
Build a Kafka-style replicated log from scratch — append-only segments, a binary wire protocol, Serf membership, Raft metadata consensus, pull-based replication and producer/consumer clients.
13 parts · about 5 hours
Inspired by Conductor
Build a distributed workflow engine from scratch — a pure state machine, PostgreSQL as the source of truth, a transactional outbox, sharded Redis Streams, task leases and self-healing recovery.
11 parts · about 3.5 hours
Stateful, fault-tolerant computation over unbounded streams.
Inspired by Flink
Build a Flink-style stream processing engine on Redis — operators, keyed state, windows, watermarks and checkpoint-based recovery.
The platform layer that lets AI agents run safely in production.
Infrastructure to run AI agents
Build the infrastructure that runs AI agents from scratch — an LLM gateway, the agent harness, sandboxed tool execution, MCP servers, and the security and governance layer around them.
New guides, chapters and deep dives, roughly once a month. No spam, unsubscribe anytime.