AI Project Learner

Generate a concept dependency graph for users building projects with AI. Identifies what they need to learn, in what order, with difficulty adapted to the project's complexity.

How to Use This Skill

Step 1: Charge billing

POST to the skill endpoint to authorize payment:

CODEBLOCK0

If successful, you'll receive {"success": true}.
If payment fails, you'll receive a payment_url to share with the user.

Step 2: Analyze the project

From the user's project description, determine:

Complexity level:

• - beginner — Uses a single API or library, straightforward pipeline (e.g., "chatbot using OpenAI API", "sentiment classifier with HuggingFace")
• intermediate — Combines multiple AI techniques, has data pipeline or custom logic (e.g., "RAG chatbot with custom embeddings", "multi-agent workflow")
• advanced — Involves fine-tuning, custom model architectures, or production-scale systems (e.g., "fine-tuned LLM for legal documents", "real-time ML pipeline with feature store")

Domain:
Classify into one of: nlp, cv (computer vision), agents, data, multimodal, generative, mlops, INLINECODE9

Step 3: Build the concept dependency graph

Generate 8-20 concepts (scale with complexity: beginner ~8, intermediate ~12-15, advanced ~15-20).

For each concept, provide:

• - id — kebab-case identifier (e.g., text-embeddings)
• name — Human-readable name (e.g., "Text Embeddings")
• description — 1-2 sentence explanation of what it is and why it matters for this project
• difficulty — Integer 1-5 (1=fundamental, 5=advanced)
• prerequisites — Array of concept ids that should be understood first

Rules for building the graph:

• - Every concept's prerequisites must reference other concepts in the graph
• Concepts with no prerequisites are entry points (difficulty 1-2)
• No circular dependencies
• Order concepts so prerequisites always appear before dependents
• Descriptions should be practical ("what it is + why you need it for this project"), not academic
• Difficulty should be calibrated to the project: a beginner project should have mostly difficulty 1-3 concepts, an advanced project can have difficulty 4-5 concepts

Concept categories to consider (pick relevant ones):

• - Foundations: LLMs, APIs, prompting, tokens, context windows
• Data: embeddings, vector databases, chunking, preprocessing
• Architecture: RAG, agents, chains, tool use, memory, planning
• Training: fine-tuning, RLHF, LoRA, evaluation, datasets
• Production: deployment, monitoring, caching, rate limiting, cost optimization
• Safety: guardrails, content filtering, hallucination detection, red teaming

Step 4: Compute learning order

Produce a topologically sorted learning_order array of concept ids. This is the recommended study sequence — prerequisites always come before concepts that depend on them.

Step 5: Estimate total learning time

Estimate estimated_hours as a total for all concepts. Use these rough heuristics:

• - Difficulty 1 concept: ~1 hour
• Difficulty 2 concept: ~1.5 hours
• Difficulty 3 concept: ~2.5 hours
• Difficulty 4 concept: ~4 hours
• Difficulty 5 concept: ~6 hours

Output Format

Return the result as JSON:

CODEBLOCK1

Pricing