Generative AI and Large Language Models: a crash course
From next-token prediction to RAG, fine-tuning, agents and deployment. Ten units and a capstone, each with a PDF and runnable code, most with interactive widgets, plus a 14-lab track. Built for a B.Tech 4th-year student.
Start with Unit 1 Open the playground Download everythingHow this course works
1 · Read
Each unit has a short web lesson and a full PDF (15-20 pages) with derivations, worked numbers and exercises.
2 · Play
The playground has one widget per big idea. Every slider has a plain label, a range and a help line, plus presets and a Reset button.
3 · Run
Every figure in every PDF comes from a script you can download and run. The core code uses only NumPy.
4 · Check
Click-to-reveal questions at the end of every lesson; each PDF has graded exercises and a recall quiz with answers.
The units
Foundations of Generative AI
Discriminative vs generative models, and the six model families: autoencoders, VAEs, GANs, autoregressive models, diffusion, Transformers.
Neural Networks and the Transformer
Why RNNs struggled, how attention fixes it, and every part of the Transformer block.
Large Language Models: Tokens, Decoding and Scaling
How an LLM reads text as tokens, predicts the next one, and how decoding settings turn probabilities into words.
Prompt Engineering and LLM Applications
Zero-shot, few-shot, chain-of-thought, structured output, tool calling, and managing context.
Retrieval-Augmented Generation
Give the model your documents: chunking, embeddings, vector search, hybrid search and how to evaluate it.
Fine-Tuning and Adaptation
Prompting vs RAG vs fine-tuning, and LoRA, QLoRA and quantization from first principles.
Multimodal and Generative Models
Diffusion, CLIP-style image-text embeddings, vision-language models and speech.
LLM Agents and Tool Use
An LLM in a loop with tools, memory and a plan; and how to keep it reliable and safe.
Evaluation, Safety and Responsible AI
Why LLMs are hard to evaluate, and how to measure and defend them.
Deployment and Industry Applications
Latency, memory, cost, quantization, serving, LLMOps and real case studies.
Capstone: a Grounded Assistant, Built, Evaluated and Deployed
Combine retrieval, evaluation and a service into one project: design brief, stage-wise diagnosis, refusal, confidence intervals, and ten project briefs.
Labs and assessment
Laboratory track
The 14 labs of the syllabus, each mapped to code you download with the units, with steps and a check-yourself question.
Assessment
A question bank with answers, a sample mid-semester and end-semester paper, and an answer key. Numbers in the answers come from the course code.
Companion course
Internet of Things and Robotics
A twelve-unit crash course built the same way: PDFs, runnable code, and one running project (a Smart Greenhouse Rover) from electronics and MQTT to kinematics, control and navigation.