Three starting paths
- Local LLMs: Memory, Quantization and Training
Begin with memory arithmetic, experiment with numerical error, then distinguish adapters from the frozen base model.
- Reliable Agents: Build, Observe and Recover
Learn the control boundaries before adding more autonomy. Follow the existing architecture and evaluation collections.
- Distributed Systems: State, Replicas and Failure
Build intuition through state changes, inspect a failure, then connect the result to a technical guide.
Evidence comes first
Each path connects explanations and explicit simulations. Measured GPU reports are published separately after real runs complete. Follow the stated prerequisites and limitations rather than treating a numerical illustration as a benchmark.