PathCraft：融合经典规划算法与大语言模型的智能职业发展系统

PathCraft is an innovative career planning system developed by ABELNoval (released on GitHub in 2026) that merges classic AI planning algorithms (STRIPS, A* search) with modern large language models (LLM). It addresses the problem of learners struggling to plan optimal learning paths by converting natural language career goals into structured paths, ensuring complete prerequisites and optimal routes—overcoming limitations of traditional systems that ignore skill dependencies.

Career planning and learning path design face four core challenges:

1. Complex prerequisite dependencies between skills, leading to learning difficulties if basics are missed.
2. Vague natural language goal descriptions (e.g., 'I want to be a data scientist').
3. Multiple paths to the same goal, requiring trade-offs between time, difficulty, and relevance.
4. Difficulty adjusting paths dynamically as industries evolve or personal progress changes.

PathCraft uses a three-layer architecture:

1. Classic Planning Layer: STRIPS formalizes state transitions (skills as predicates, actions as prerequisites/effects); A* search finds optimal paths using heuristics (course difficulty, duration, career relevance).
2. LLM Layer: Handles intent understanding (extracts skills from goals), course matching (maps skills to database entries), path explanation (human-readable notes), and content generation (supplementary materials).
3. Data Layer: Relies on skills_library.json (skill dependencies/difficulty) and courses_db.json (course details, required skills, duration) for flexibility across scenarios.

Users interact via CLI (e.g., python src/main.py --current_skills 'Python, Basic Math' --target 'Data Scientist'). Steps:

1. Goal Parsing: LLM extracts key skills (statistics, ML, data visualization) from the target.
2. Gap Analysis: Compares current skills with target skills to identify gaps.
3. Topological Sort: Orders skills by dependency constraints.
4. Path Search: A* algorithm finds optimal course sequences.
5. Result Output: Provides course order, estimated duration, and learning suggestions.

Advantages:

• Interpretability: Clear logic for path steps (why A before B).
• Constraint Guarantee: Strictly adheres to prerequisites (unlike pure LLM paths).
• Flexibility: Handles natural language input without specific query languages.
• Scalability: Modular components allow independent upgrades.

Applications: Online education platforms (personalized paths), enterprise training (skill gap solutions), career consulting (transition plans), higher education (course planning).

Limitations: Manual data maintenance, static paths, single optimization metric (path length), no user preference modeling. Future Plans: Auto extract structured data from course descriptions, dynamic path adjustment, multi-objective optimization (cost/time flexibility), user learning style modeling.

Conclusion: PathCraft showcases the value of combining symbolic AI (rigor, interpretability) with neural models (flexibility, natural language handling) to solve real-world career planning problems, offering a balanced and practical solution.