Generative AI
Generative AI end to end: LLMs, prompting, fine-tuning, RAG, agents, diffusion, VLMs and running GenAI in production.
What Is Generative AI?
The field that creates text, images, audio and code — and the system stack behind it.
The Mathematics of Generation
Generation is sampling from a probability distribution over tokens — conditioned on context.
Text Generation Fundamentals
Generate text with pretrained models: sampling strategies, decoding, and the tokens that come out.
Prompt Engineering
Design instructions that get reliable outputs: roles, structure, few-shot examples and constraints.
In-Context Learning
Teach new tasks with examples in the prompt — no weight updates required.
LLM Architecture & Scaling
Decoder-only transformers, scaling laws, and why bigger models behave differently.
Tokenization & the Vocabulary
Subword tokens (BPE) — the hidden layer between characters and embeddings.
Fine-Tuning LLMs
Adapt a foundation model to a domain with supervised fine-tuning and parameter-efficient methods (LoRA).
RLHF & Alignment
Train models to be helpful and harmless: reward models, RLHF and constitutional methods.
Retrieval-Augmented Generation (RAG)
Ground answers in your own data: retrieve relevant chunks, stuff the context, generate.
Embeddings & Vector Databases
Semantic search at scale: embedding models, vector indexes, and ANN search.
Agents & Tool Use
Let models call functions, use tools, and loop until the job is done.
Diffusion Models for Images
Generate images by learning to denoise: forward noise, reverse denoising, and text conditioning.
Generative Adversarial Networks
Two networks in a zero-sum game: the generator fools the discriminator, and both improve.
Vision-Language Models
Bridge images and text: CLIP, captioning, and multimodal chat (LLaVA, GPT-4V).
Generative Audio & Speech
Text-to-speech, voice cloning and music generation with modern neural audio models.
Evaluating LLMs
Benchmarks, evals and human review — measuring quality, safety and reliability.
LLMOps: Running GenAI in Production
Prompt management, caching, guardrails, monitoring and cost control for LLM apps.
Ethical AI & Safety
Bias, privacy, transparency and misuse — the responsibilities that come with generative models.
Building a GenAI Application
Assemble the stack: model, retrieval, guardrails and a clean interface — end to end.
Generative AI Roadmap
Synthesize everything into a career and project plan — and keep up as the field accelerates.