AI 🤖
Artificial Intelligence, Large Language Models (LLMs), Agentic Architectures, and Generative AI systems.
1. Multi-Agent Systems & Orchestration
Production-grade curricula on agentic workflows and LLM orchestration:
🕸️ LangGraph Architecture Curriculum
Comprehensive 12-part deep dive on cyclic graphs, persistent memory, and agent orchestration:
- 01 — Mental Model — StateGraph, nodes, and edges
- 02 — State & Reducers — Schema definition, channels, and operator reducers
- 03 — Nodes & Edges — Conditional routing, START, and END nodes
- 04 — Tools & Routing —
ToolNode, tool calling, and dynamic routing - 05 — Command & Interrupts — Dynamic graph mutations and control flow
- 06 — Subgraphs — Hierarchical multi-agent compositions
- 07 — Streaming — Token streaming (
messagesmode) and state updates (valuesmode) - 08 — Checkpointers & Persistence — MemorySaver, Postgres, and time-travel debugging
- 09 — Memory Store — Cross-thread memory and semantic recall
- 10 — Human-in-the-Loop — Breakpoints, approvals, and state edits
- 11 — Production Deployment — LangGraph Cloud, async queues, and resilience
- 12 — Testing & Evaluation — Unit testing agents and deterministic fixtures
🦜 LangChain Core Framework
- 01 — Mental Model — Chains, components, and LCEL
- 02 — Message Architecture — System, Human, AI, and Tool messages
- 03 — Chat Models — Providers, parameters, and invocation
- 04 — Tool Calling — Pydantic tool definitions and validation
- 05 — Prompt Templates — ChatPromptTemplate and message placeholders
- 06 — Runnables & LCEL — Composition with
|, batching, and parallel execution - 07 — Memory & Callbacks — Observability hooks and tracing
- 08 — LangGraph Intro — Transition from LCEL chains to cyclic graphs
- 09 — Streaming — SSE streaming events
- 10 — Testing — Unit testing LCEL chains
2. Core Concepts & Inner Workings
- AI Inner Workings Hub — How transformers and foundation models work
- AI vs ML vs GenAI — Core distinctions and taxonomy
- What is the ‘T’ in GPT? — Self-attention, multi-head attention, and transformer encoders/decoders
- Embeddings & Vector Representations — High-dimensional latent space and cosine similarity
- Fine-Tuning — SFT, LoRA, QLoRA, and parameter-efficient tuning
- RLHF & Alignment — Reward models and PPO/DPO
- Prompt Engineering Patterns — Few-shot, chain-of-thought, ReAct, and system instructions
3. RAG, Agents & Standards
- AI Agents — The agent loop, tool design, workflows versus agents, context management, guardrails
- Evaluating LLM Systems — Datasets, code and model graders, offline and online evaluation, metrics for RAG and agents
- Retrieval-Augmented Generation (RAG) — Advanced RAG pipelines, chunking, hybrid search, and re-ranking
- RAG in Production — Ingestion, freshness, access control, diagnosing bad answers, cost and latency
- Vector Databases — ANN indexes, quantization, filtering, hybrid search, pgvector versus dedicated engines
- Model Context Protocol (MCP) — Standardized client-server protocol for connecting AI agents to tools and data
- MCP in Production — Transports, OAuth authorization, tool design, security threats, deployment
- Prompt Injection & Jailbreaks — Security risks, indirect injection, and mitigation guardrails
4. Local AI & Cloud Infrastructure
- Running AI Locally (Ollama) — Local inference, quantization (GGUF), model pulling, and GPU acceleration
- AWS AI Services Hub — Cloud machine learning and generative AI
- Amazon SageMaker — Managed training, endpoints, and inference options
- Amazon Bedrock — Foundation models, Knowledge Bases, and Guardrails
Related Knowledge Bases
- ☸️ Kubernetes GPU Scheduling — DRA, GPU node pools, and vLLM on K8s
- ☁️ AWS GPU Compute — P4d/P5 instances, Trainium, and Inferentia
- 🔷 Azure AKS GPU Infrastructure — InfiniBand, NVIDIA H100/H200, and KubeRay
- 🐧 Linux Kernel & Drivers — NVIDIA CUDA drivers, NUMA, and sysctl tuning