AI Learning Path: From Fundamentals to Building Next-Gen Tools

We’re building AI learning tools incrementally — starting with fundamentals and scaling to advanced concepts. This is the curriculum we’re following.

The Learning Path

Level 1: Fundamentals

Goal: Understand how LLMs work and how to prompt them effectively.

Concepts:

• Tokenization and context windows
• Temperature, top-p, top-k parameters
• System prompts vs user prompts
• Few-shot and chain-of-thought prompting
• Structured outputs (JSON mode)

Tool: prompt-lab — Interactive playground for testing prompts across models

Deliverables:

• Prompt comparison tool (same prompt, different models)
• Parameter sensitivity analyzer
• Token counter and cost estimator
• Prompt template library

---

Level 2: Prompt Engineering

Goal: Master techniques for reliable, high-quality outputs.

Concepts:

• Role prompting and persona injection
• Decomposition (breaking complex tasks into steps)
• Self-critique and refinement
• Negative prompting (what NOT to do)
• Prompt chaining and sequencing

Tool: prompt-engineer — Multi-step prompt builder with evaluation

Deliverables:

• Chain builder (visual prompt sequencing)
• Output quality scorer
• A/B testing framework for prompts
• Prompt version control

---

Level 3: Retrieval-Augmented Generation (RAG)

Goal: Build systems that combine LLMs with external knowledge.

Concepts:

• Document chunking strategies
• Embedding models and vector databases
• Similarity search (cosine, dot product, hybrid)
• Reranking and context optimization
• Hallucination detection

Tool: rag-builder — Visual RAG pipeline builder

Deliverables:

• Chunking strategy comparison tool
• Embedding model benchmark
• RAG evaluation framework (faithfulness, relevance)
• Vector DB comparison (Pinecone, Weaviate, Chroma, etc.)

---

Level 4: Fine-Tuning

Goal: Customize models for specific tasks and domains.

Concepts:

• When to fine-tune vs prompt engineering
• Data preparation and formatting
• LoRA, QLoRA, and full fine-tuning
• Evaluation and overfitting detection
• Deployment and serving

Tool: finetune-lab — Fine-tuning experiment tracker

Deliverables:

• Data quality checker
• Training progress visualizer
• Model comparison (base vs fine-tuned)
• Cost/performance tradeoff analyzer

---

Level 5: Multi-Agent Systems

Goal: Build systems where multiple AI agents collaborate.

Concepts:

• Agent architectures (ReAct, plan-and-execute)
• Communication protocols between agents
• Task decomposition and delegation
• Consensus and conflict monitoring
• Emergent behavior

Tool: agent-orchestrator — Multi-agent workflow builder

Deliverables:

• Agent communication visualizer
• Task decomposition analyzer
• Consensus mechanism comparison
• Multi-agent benchmark suite

---

Level 6: Production Systems

Goal: Deploy AI applications at scale with reliability.

Concepts:

• Caching strategies (exact, semantic, prompt)
• Rate limiting and queue management
• Fallback chains and error handling
• Cost optimization and monitoring
• A/B testing in production

Tool: production-monitor — AI system observability platform

Deliverables:

• Cost tracking dashboard
• Latency monitoring
• Quality regression detection
• Automatic failover testing

---

Level 7: Advanced Architectures

Goal: Build next-generation AI applications.

Concepts:

• Mixture of Experts (MoE) routing
• Speculative decoding
• Tree of thoughts
• Constitutional AI
• Recursive self-improvement

Tool: architect-lab — Advanced pattern playground

Deliverables:

• MoE routing visualizer
• Speculative decoding demo
• Tree of thoughts implementation
• Constitutional AI testing framework

---

Level 8: Building Products

Goal: Ship AI-powered products that users love.

Concepts:

• User experience design for AI
• Feedback loops and continuous improvement
• Ethical AI and bias mitigation
• Scaling and infrastructure
• Business models for AI products

Tool: product-builder — AI product development framework

Deliverables:

• User feedback collection system
• Bias detection toolkit
• Scaling cost projector
• AI product teardown library

---

Progress Tracking

Completed

• Level 0: Tool reviews (Hermes, Claude Code, Cursor, etc.)
• Level 0: Provider analysis (OpenRouter, Together, etc.)
• Level 0: Benchmark framework (8 models tested)

In Progress

• Level 1: prompt-lab fundamentals tool
• Level 1: Prompt comparison across models

Upcoming

• Level 2: prompt-engineer chain builder
• Level 3: rag-builder pipeline tool
• Level 4: finetune-lab experiment tracker

How to Follow Along

1. Start at Level 1 — even if you’re experienced, fundamentals matter
2. Build the tools — each level has hands-on deliverables
3. Share your results — community learning is faster
4. Iterate — go back to earlier levels as you learn more

Contributing

We’re building in public. Contributions welcome:

• Code: Build the tools alongside us
• Content: Write tutorials for each level
• Testing: Use the tools and report issues
• Ideas: Suggest new levels or concepts

The goal: the most comprehensive AI learning resource on the web, built one level at a time.