A pattern-first course on LeetCode and coding-interview mastery, in Python. You learn the recurring patterns, not a pile of memorized solutions, and you build the recognition reflex that works under interview pressure with no AI in the room.
Muster covers the algorithmic and data-structure round: pattern recognition under time. It does not cover system design, behavioral, SQL, or object-oriented-design rounds; the Week 16 continuation plan points you to those next.
Muster is fronted by a demanding Socratic tutor that runs inside your AI coding agent. The tutor does not write your solutions, name the pattern before you have worked for it, or grade your code by reading it. It enforces the discipline that turns sixteen weeks of practice into a reflex you can perform in an interview.
Muster is a Lernen-edu course, sibling to Vigil (cybersecurity) and Agentwright (agentic engineering).
# 1. Clone the course
git clone https://github.com/lernen-edu/muster.git
cd muster
# 2. Open it in your AI coding agent (Claude Code shown; see the support matrix below)
claude
# 3. Ask the tutor to orient you
# "What are you, and what are the rules?"
Then read, in order: getting-started.md (set up Python and your agent, and work the Python refresher), AI-ETHICS.md (why this course keeps AI out of the problem-solving), and curriculum/week-00-setup/README.md (your first week).
The course assumes you have written a little Python but treats you as not yet fluent. Week 0 includes a refresher and a self-check that gates Week 1, and every week refreshes the idioms it leans on.
A working adult finishes Muster in about 16 weeks at 12 to 15 hours per week. Full-time pacing compresses it to roughly 10.
Muster enforces five pedagogical principles. They are rules, not suggestions, and the tutor holds you to them:
- Pattern first, code second. Before the editor opens, you name the pattern aloud and write the approach in plain English as a comment.
- The struggle floor: 25 minutes. No hints, no solutions, no AI, no Discuss tab, for the first 25 minutes of every new problem.
- Hints, not solutions. A six-rung hint ladder, one rung per ask, maximum six. The tutor never hands you the code.
- Debrief every problem. A five-question debrief committed in the commit message before the next problem unlocks.
- Spaced repetition is the curriculum. Every Friday, two prior-week problems re-solved cold, no notes.
The full human-facing contract is in tutor-reference.md. The tutor's enforcement contract is .tutor/tutor-core.md.
A repository of problems you solved yourself, each commit carrying a debrief that names the pattern, the trigger, the complexity, and the edge case that nearly broke you; a set of Pattern Cards in your own words, one per family; and a RETROSPECTIVE.md of strong and weak patterns with a continuation plan. The real artifact is the reflex: read an unfamiliar problem, feel its family, reach for the right structure, all with no model in the room.
Full master index and per-week detail: curriculum/README.md.
| Week | Pattern | Canonical deep-dive |
|---|---|---|
| 0 | Setup and Python refresher | (n/a) |
| 1 | Arrays and Hashing | Two Sum |
| 2 | Two Pointers | 3Sum |
| 3 | Sliding Window | Minimum Window Substring |
| 4 | Stacks | Largest Rectangle in Histogram |
| 5 | Binary Search | Koko Eating Bananas |
| 6 | Linked List | Reorder List |
| 7 | Trees | Binary Tree Maximum Path Sum |
| 8 | Heap / Priority Queue (+ consolidation) | Find Median from Data Stream |
| 9 | Backtracking | N-Queens |
| 10 | Tries and Intervals | Word Search II |
| 11 | Graphs | Course Schedule |
| 12 | Advanced Graphs | Cheapest Flights Within K Stops |
| 13 | 1-D Dynamic Programming | Coin Change |
| 14 | 2-D Dynamic Programming | Edit Distance |
| 15 | Greedy, Bit, Math and Geometry | (multiple) |
| 16 | Mock Interview Marathon | (cross-pattern) |
The tutor activates from a per-agent entry-point file. Pick one agent and commit to it for the course.
| Agent | Entry-point file | Mode |
|---|---|---|
| Claude Code | CLAUDE.md |
Full enforcement |
| Codex CLI | AGENTS.md |
Full enforcement |
| Gemini CLI | GEMINI.md |
Full enforcement |
| OpenCode | AGENTS.md |
Full enforcement |
| Pi Coding Agent | AGENTS.md |
Full enforcement |
| GitHub Copilot | .github/copilot-instructions.md |
Advisory (see the file) |
Each file redirects the agent to .tutor/tutor-core.md. There is no install script and no MCP server; the entry-point files plus the tutor core are the entire distribution mechanism.
muster/
├── README.md # This file
├── getting-started.md # Python + agent setup, and the Python refresher
├── tutor-reference.md # The human-facing tutor contract
├── AI-ETHICS.md # AI-use ethics + interview integrity (read before Week 1)
├── CLAUDE.md AGENTS.md GEMINI.md # Per-agent tutor entry points
├── .github/copilot-instructions.md # Copilot entry point
├── .tutor/ # Tutor contract + runtime state (not part of the public site)
│ ├── tutor-core.md # The tutor's enforcement contract (the source of truth)
│ ├── session.json # Per-learner session state (template)
│ ├── progress.md revisit-log.md
│ └── pattern-cards/ # Your pattern cards live here
└── curriculum/ # The course
├── README.md # Master 16-week index
└── week-00 .. week-16/
The tutor is demanding by design. It will not say "great question." It will not soften a refusal because you are frustrated. It will not write your solutions, and it will not apologize for any of that. If you want a warm assistant, there are many; this is not one. The bluntness is the point: interviews are performed under pressure, and the muscle for it is built here.
This course keeps AI out of the problem-solving on purpose, because the interview you are training for does too. Using AI to solve practice problems, or to look further along than you are, only sets up a failure in the room where it counts. Read AI-ETHICS.md before you begin.
Dual-licensed:
- Course content (everything under
curriculum/, plusindex.md,getting-started.md,tutor-reference.md, andAI-ETHICS.md) is licensed under CC BY 4.0. - Tutor scaffolding and code (
.tutor/, the per-agent entry-point files, and_config.yml) is licensed under the MIT License, copyright Lernen-edu.
See LICENSE-CONTENT and LICENSE-CODE for the full terms.