Guides

Curated routes through the atlas — ordered paths that thread many explainers into one journey.

• The System Design Interview Ramp — The data-infra loop, end to end — foundations, the storage-streaming-query core, concurrency & Python internals, low-level & object-oriented design, interview craft, and timed rehearsal. Learn by operating the systems, not memorizing boxes.
• General System Design — System design beyond the data-infra core, now in themed phases — the method on one page, then feeds and messaging, search and the world, media and files, commerce and money, markets and realtime. Foundations on the data-infra ramp, breadth here.
• The AI Interview Ramp — A phased path through the AI/ML atlas, forking into two lanes — Research Scientist (evals, optimization, scaling, post-training) and Staff MLE / AI Infrastructure (serving, training at scale, checkpointing, RAG). Foundations shared, depth by lane.
• The Agent-Infrastructure Interview Ramp — A phased reading route for a tool-calling, webhook, and sandboxed-execution interview — the method and mechanics as explainers, then eight foundational papers on research.rudrite.com, in the order the field actually built them.
• The JAX Stack, From the Inside — A bottom-up path through the modern JAX training/serving stack, for reading real open-source code rather than interview prep — jax, flax, orbax, xla, qwix, maxtext — tracing to sharding to checkpointing to quantization to production.
• PyTorch/XLA, From the Inside — A project track: PyTorch/XLA read from a Colab notebook down to the PJRT call. Notebooks you run on a TPU, the lazy tensor and its compile cache, then the seam between torch_xla and XLA read line by line on both sides, every file pinned to one commit.
• XLA, From the Inside — A project track: openxla/xla read from a dump on disk down to the thunk that runs. PJRT and StableHLO ingestion, HLO and the pass runner, fusion, buffer and layout assignment, SPMD and collectives, the CPU runtime, GPU codegen, IFRT, every file pinned to one commit.
• Pallas, From the Inside — A project track: the kernel language inside jax and the tokamax library built on it. A kernel you run in interpret mode, pallas_call and BlockSpecs, the TPU pipeline and both GPU lowerings, flash and splash attention as real kernels, and an autotuned Op.
• JAX Core, From the Inside — A project track inside jax-ml/jax: a jaxpr you print, the tracer and partial evaluator that built it, the jit cache key, autodiff and batching as rule tables, pytrees and control flow, shard_map and the lowering to StableHLO, every file pinned to one commit.
• Tunix, From the Inside — A project track: JAX-native post-training read from a GRPO recipe down to the rollout worker. Upstream notebooks you run, RLCluster and its meshes, the learner loop, the sampler and rollout backends with weight sync, agentic trajectories, and the PEFT trainer.
• Line by Line — the Code-Walk Shelf — Every code-walk on the site, one shelf — real production files read whole and byte-exact: CPython’s stdlib, a write-ahead log, Redux, and the training-stack internals. Derived from the catalog, so a new walk joins the day it ships.
• Python, Read From the Source — A phased path through CPython and the standard library — the interpreter’s own data structures, three stdlib files read line by line, the evaluation-model gotchas, and the craft of writing Python that reads the same way.