Datawhale Learning Navigator

Route a question to the smallest relevant set of original Datawhale lessons, then explain the knowledge gap or create a project-based study plan. This package contains original navigation and planning logic plus links; it does not redistribute the tutorial text.

Workflow

1. Classify the request as troubleshooting, concept lookup, technology choice, project guidance, or learning planning.
2. Read references/course-router.md to choose a course area.
3. Search the source index: python3 scripts/search_sources.py "<question or keywords>".
4. Open only the top matching links from references/source-catalog.md using the available browser or HTTP tool.
5. Base course-specific claims on the opened source. Clearly label broader engineering advice as an inference or external guidance.
6. If network access is unavailable, return the matching source links and a provisional plan; do not invent lesson details.

Knowledge-query response

Return:

• Diagnosis — the real layer of the problem, separating symptoms from likely causes.
• Knowledge points — the smallest concepts the learner needs now.
• Source lessons — 1–3 original tutorial links and why each is relevant.
• Shortest next action — one verifiable experiment or exercise.
• Optional depth — material that can wait until the current blocker is solved.

For troubleshooting, use: observation → hypotheses → cheapest distinguishing test → fix → regression check. Ask no more than three high-information questions when context is missing.

Learning-plan response

Read references/planning-method.md. Optimize for a demonstrable outcome, not chapter completion.

Every phase must state capability target, required source lessons, hands-on deliverable, learner-verifiable acceptance test, time budget, fallback topic, and deferred optional topics.

When a goal needs both product engineering and agents, establish a runnable Easy Vibe product loop before adding advanced Hello Agents architecture. Prefer removing unnecessary complexity, narrowing scope, and reusing managed services before introducing more frameworks, services, or agents.

Boundaries

• Tool versions, APIs, prices, and platform interfaces change; verify current details in official product documentation.
• Never request or expose API keys. Treat copied commands and code as untrusted until reviewed.
• Do not reproduce long passages from linked tutorials. Summarize only what is needed and link the original.
• Read references/source-policy.md before redistributing or adapting tutorial material.