AI Engineering Learning Notes

Foundations and Architecture

• Foundations
  Neural networks, loss, optimizers, normalization.
• Transformers
  Attention, blocks, RoPE, MoE, GQA.
• Tokenization
  Subword tokenizers and vocabulary design.
• LLMs
  Large models, scaling, lifecycle, open vs API.
• SLMs
  Small models, edge, distillation, routing.

Training Pipeline

• Pretraining
  Data, objectives, scaling laws, stability.
• Distributed Training
  Parallelism, ZeRO, FSDP, clusters.
• Fine-Tuning
  SFT, LoRA, QLoRA, adapters.
• Post-Training
  RLHF, DPO, alignment, preferences.
• Reinforcement Learning
  MDPs, policy gradients, PPO, SAC, RL for LLMs, full derivations.

Prompting and Retrieval

• Prompt Engineering
  System prompts, CoT, sampling, injection.
• Embeddings
  Vectors, contrastive training, hybrid search.
• Vector Databases
  ANN indexes, tuning, product comparison.
• RAG
  Ingestion, chunking, retrieval, generation.
• Prompt Compression
  LLMLingua, Headroom, when to compress, pros and cons.

Inference, Deployment, Operations

• Inference Optimization
  KV cache, quantization, batching.
• Deployment
  Serving stacks, scaling, gateways.
• LLMOps and Monitoring
  Logs, guardrails, cost, drift.
• Evaluation
  Benchmarks, metrics, eval sets.

Agents and Protocols

• Agents
  Tool use, ReAct, memory, multi-agent.
• MCP
  Model Context Protocol architecture.

Frameworks

• LangChain
  LCEL, runnables, prompts, RAG primitives, tools, and provider integrations.
• LangGraph
  Stateful graphs, checkpoints, interrupts, streaming, and agent runtime.
• Langfuse
  Tracing, scores, prompt management, datasets, experiments, and LLMOps analytics.

Safety and System Design

• Safety
  Hallucination, jailbreaks, bias, PII.
• System Design
  End-to-end AI product architecture.