Kafka Queues and Share Groups
Notes from Jun Rao on Queues for Kafka: share groups, share partitions, per-record delivery state, and how Kafka evolves without abandoning the log.
Technical essays, paper notes, and implementation writeups about AI, Databases and anything remotely interesting!
A practical guide to improving AI coding systems by fixing the instructions, tools, environment, state, and feedback loops around the model.
Notes from Jun Rao on Queues for Kafka: share groups, share partitions, per-record delivery state, and how Kafka evolves without abandoning the log.
Context engineering is the natural progression of prompt engineering — not about writing better prompts, but about curating exactly which information enters the limited context window.
Notes on LeWorldModel, JEPA-style world models, latent planning, physical understanding, and why world models matter for AI agents.
Notes on Kafka's original design: logs, partitions, consumer groups, ZooKeeper coordination, and throughput-oriented tradeoffs.
Notes on harnesses and evaluation loops for making model-assisted systems easier to inspect.
Notes on how ToolFormer taught language models when and how to call external tools from self-supervised data.
How I approached region-level locking, leases, and commit validation for safer multi-agent Python editing.