LLMs and GenAI¶

Everything in the repo on language models, retrieval, agents, fine-tuning, evaluation, deployment, and the surrounding tooling. Start here whether you're trying to understand transformers, ship a RAG system, pick a vector database, or pass an AI cert.

flowchart LR
  L[Learn: concepts] --> B[Build: hands-on]
  L --> C[Compare: services]
  C --> B
  B --> Cert[Certify: AI certs]
  L --> Cert

Learn¶

Plain-English concept pages, 5-10 minute reads:

Foundations - LLM basics - what an LLM actually is, tokens, context, why it can't count - Transformer architecture - attention, self-attention, what's inside the box - Embeddings and vector search - dense representations, similarity - Context windows and management - tokenization, limits, summarization, retrieval as extension - Multimodal models - vision, audio, video inputs

Building with LLMs - Prompt engineering - the patterns that work, the ones that don't - Tool use and function calling - giving an LLM hands - MCP explained - Model Context Protocol, the standard interface for tools - Structured outputs - JSON mode, schemas, constrained decoding - RAG explained - retrieval-augmented generation, the workhorse pattern - Fine-tuning vs RAG - when to use which (mostly RAG) - Agents explained - tool-using LLMs in a loop - Agentic loops - ReAct, planner/executor splits, harnesses, failure modes - Prompt caching - cache hits, when caching pays off

Operations - Evals for LLMs - how to know if changes helped or hurt - Guardrails and safety - input/output filtering, content moderation - Inference servers - vLLM, TGI, SGLang, llama.cpp - Quantization and distillation - smaller models, less memory

Compare¶

Cross-vendor comparisons:

• GenAI platforms - Anthropic API, OpenAI, AWS Bedrock, Azure OpenAI, Vertex AI, Together, Fireworks, Groq
• Vector databases - Pinecone, Weaviate, Qdrant, Milvus, pgvector, OpenSearch, Azure AI Search, Vertex Vector Search, Bedrock Knowledge Bases
• Agent frameworks - Claude Agent SDK, LangGraph, CrewAI, Autogen, OpenAI Agents SDK
• LLM observability - LangSmith, Langfuse, Helicone, Phoenix, Braintrust
• AI/ML services (cloud-native) - SageMaker vs Vertex AI vs Azure ML

Reference¶

• Architecture pattern: AI/ML pipeline - production deployment

Build¶

Hands-on projects with inline code:

• Build a RAG pipeline - load, chunk, embed, retrieve, generate, eval
• Build a Claude agent with MCP - tool-using agent with a custom MCP server
• Run Llama on a single GPU - vLLM, OpenAI-compatible endpoint
• Set up an eval harness - golden set, regression detection, CI integration
• Fine-tune with LoRA - LoRA on a small open model, eval against base

Certify¶

Certs and study tracks that cover LLMs and GenAI:

Vendor study tracks (no formal exam) - Anthropic Claude Architect Foundations - Anthropic Claude Architect Professional - Anthropic Claude Developer Foundations - Anthropic Claude Prompt Engineering Specialist

Foundational - AWS AI Practitioner - cross-cert GenAI study track - Azure AI Fundamentals (AI-900)

Associate - AWS ML Engineer (MLA-C01) - Azure AI Engineer (AI-102) - GCP Machine Learning Engineer - Databricks Data Engineer Associate - feeds GenAI work - Databricks GenAI Engineer Associate - Databricks ML Associate - NVIDIA GenAI/LLM Associate - NVIDIA Multimodal GenAI Associate - NVIDIA AI Infrastructure & Operations Associate

Professional - Databricks ML Professional - NVIDIA GenAI/LLMs Professional - NVIDIA Agentic AI Professional - NVIDIA AI Infrastructure Professional - NVIDIA AI Operations Professional - NVIDIA Accelerated Data Science Professional

Roadmap¶

The career-track view of these certs lives in AI/ML Engineer roadmap.