One climb, beginner → architect. Every rung is a real role with its own skills and an XP gate. Clear a stage's lessons and pass its gate to be promoted to the next.
Prove you can make a computer do something in three minutes. Print, variables, f-strings — the run-it-and-see-it dopamine hit.
Control flow, booleans, loops, debugging — the muscle memory of code that branches and repeats. Ends on your first boss build.
Functions, scope, and data structures — lists, dicts, tuples, sets, comprehensions. Your first multi-feature project a stranger could use.
Files & JSON, the stdlib, calling an API, secrets & env-vars, error handling, OOP and your first pytest. Code that survives failure and talks to services.
Your first specialism. Schema & batch validation, feature engineering, a rule-based classifier, evaluation metrics, a train→predict→evaluate loop — hand-rolled pure Python.
The advanced pure-Python patterns that separate mid from senior — generators, decorators, context managers, asyncio basics, plugin registries, state machines.
LLM-from-code fundamentals — prompting, system prompts & memory, calling an LLM (stubbed), tool-call dispatch, JSON guardrails. Ends on the AI-Assistant core boss.
Production ops patterns — retry with backoff, idempotency, config-driven dispatch, rate-limiting, dead-letter handling, step pipelines, scheduling. The bridge into AI engineering.
Build real AI features — few-shot & structured prompting, robust function-calling, RAG with embeddings + vector search, a minimal agent loop, guardrails, and evals.
The systems view — multi-agent orchestration, vector-DB design, pipeline/DAG design, scaling & batching, cost control & token budgeting, observability, failure-mode design.
The durable human skill — reviewing other people's code. Read a diff and find the bug, spot the missing edge case, judge structure, catch the security issue, give good feedback.
The headline wedge — reviewing and verifying AI-generated code. Spot the confident-but-wrong output, the hallucinated API, the plausible-but-insecure pattern; write the harness that proves it; decide accept/reject/revise.