136 topics is a lot. These paths cut across the existing sections and give you a
focused trail based on what you're trying to do — not what section something lives in.
Your original sections are untouched. These paths are an overlay.
⚡ Quick Start
Mental models before anything else. You don't need to understand transformers to
build with AI — but you do need to know what a token is, why context windows matter,
and what an embedding actually does.
~2 days. 5 of these 7 are already written.
🧱 Essentials
The unavoidable core. If you're building anything serious with AI — apps,
integrations, agents — these are the topics you'll hit walls without.
~2 weeks. Prerequisite: Quick Start.
Prompting & Control
RAG & Retrieval
Agents
Safety Baseline
Eval Basics
🔨 For Builders
Everything you need to actually build and integrate AI into products — APIs,
streaming, agents, observability, and how to connect AI to your existing stack.
~3 weeks. Prerequisite: Essentials.
Integration Layer
Agents & Orchestration (Builder Focus)
Observability for Builders
Eval for Builders
New Patterns
⚙️ Systems Depth
The remaining ~79 topics from the original map. These don't go away — they're the
ones that separate engineers who use AI from engineers who design AI systems.
Work through them in the original section order once you've completed For Builders.
- 01 Model Inference Core (remaining 10) — KV cache, continuous batching, paged attention, FlashAttention, chunked prefill, speculative decoding, MoE, multimodal LLMs, reasoning models
- 02 Model Optimization (all 11) — Quantization, FP8/INT8/INT4, GPTQ/AWQ/GGUF, pruning, distillation, LoRA/QLoRA, adapters, fine-tuning/SFT, RLHF, DPO/GRPO, model merging
- 03 Serving Infrastructure (all 15) — vLLM, TGI, TensorRT-LLM, SGLang, llama.cpp/Ollama, Triton, P/D disaggregation, tensor/pipeline/expert parallelism, serving metrics, batch inference, edge inference, NVIDIA Dynamo
- 04 Retrieval & Memory (remaining 6) — Semantic search, knowledge graph, GraphRAG, long-context retrieval
- 05 Agents & Orchestration (remaining 7) — Plan & execute, Agent SDK, multi-agent systems, handoff A2A, A2A protocol
- 06 Prompting & Control (remaining 3) — In-context learning, constitutional AI, system card
- 07 Evaluation & Quality (remaining 9) — Faithfulness metrics, DeepEval, benchmark evals, regression testing, trajectory eval, CI/CD gates, PROMOTE/HOLD/ROLLBACK
- 08 Observability & Ops (remaining 10) — OpenTelemetry, latency percentiles, prompt/semantic drift, toxicity scoring, RBAC/IAM, audit logs, Arize/Datadog, data residency, model versioning, spans & traces
- 09 Integration & Cloud (remaining 7) — KV cache-aware routing, prefix-aware routing, cloud platforms, Hugging Face endpoints, data sovereignty, VPC/private endpoints, on-prem/hybrid
- 10 Safety & Governance (remaining 12) — Content filtering, red-teaming, adversarial inputs, safety alignment, bias detection, fairness metrics, human approval gate, policy-as-code, AI RMF, EU AI Act, responsible AI