AI and LLM Engineering

Artificial Intelligence, especially Generative AI and Large Language Models has rapidly moved from research labs into production systems and everyday tools. Yet much of the learning around AI today is driven by hype, tool-chasing, and shallow intuitions .

This course is designed to do the opposite.

is a thinking-first, engineering-oriented course that focuses on:

• correct mental models,
• system-level understanding,
• and disciplined engineering practices.

The goal is not merely to use AI tools, but to understand what they are, how they work, and where they fail .

This course is not :

• a prompt-hacking workshop
• a tool-centric crash course
• a shortcut to “AI mastery”
• a collection of disconnected demos

Instead, it is a structured journey into understanding AI systems as engineered artifacts.

This course is suitable for:

• Students who want a strong conceptual foundation in AI
• Working professionals looking to upskill responsibly in AI/LLMs
• Engineers who want to go beyond APIs and prompts
• Anyone skeptical of AI hype and interested in clear thinking

No prior deep learning expertise is required, but basic programming familiarity is expected.

I am Pa Mu Selvakumar, a Senior Principal ML Research Engineer at Saama AI Research Lab. My work spans: deep learning, reasoning systems, language understanding, interpretability and NLP. I have publications in exploring attention, graph neural networks sequence models for biomedical NLP, interpretability in question answering, machine learning approaches to differential equations and my favorite Tamil language computing

Links:

• Blog: https://pori.vanangamudi.org/posts
• Google Scholar: https://scholar.google.com/citations?user=\_KPtoVcAAAAJ&hl=en
• GitHub: https://github.com/vanangamudi
• GitLab: https://gitlab.com/vanangamudi
• LinkedIn: https://www.linkedin.com/in/selvapm/

AI is not a single technology. It is:

• a field of study spanning computer science, psychology, cognitive science, neuroscience, and philosophy
• a system or programs that appear intelligent
• an ever-moving goalpost that reshapes our definition of intelligence

Understanding this broader context is essential for any serious AI engineer.

Modern generative models can produce artifacts across multiple domains:

• Text: coding assistance, summarization, question answering
• Images: generation, editing, transformation
• Audio: speech recognition, speech synthesis, music generation

But capability alone does not imply understanding, reasoning, intention, or intelligence. Distinguishing these is a recurring theme throughout the course. Key questions explored in the course include: (i) is language competence equivalent to intelligence? (ii) do LLMs understand or simulate understanding? (iii) how should intelligence be tested?

We examine classical and modern perspectives, including:

• The Turing Test
• Embodied intelligence thought experiments
• The Chinese Room argument

These discussions are not abstract philosophy, they directly inform how we engineer and evaluate AI systems.

One of the central ideas in modern AI is the shift from: encoding tasks in architecture to encoding tasks in textual input. The following are the broad areas we will cover in this course.

• Fundamentals
  • Embeddings and vector representations
  • Mathematical intuitions
  • Neural networks
  • Datasets
  • Training vs inference
• Prompting
  • Prompt structure
  • Chain-of-thought reasoning
  • Limitations of prompting
• Retrieval-Augmented Generation (RAG)
• Fine-Tuning
  • Parameter-Efficient Fine-Tuning (PEFT)
  • Full fine-tuning
  • Data curation strategies
• Agents and Agentic AI
  • Tool use
  • Planning
  • Feedback loops
  • Failure modes
  • Learning Philosophy

• Conceptual lectures
• Live demonstrations (including AI-assisted coding)
• Hands-on exercises
• Open discussions and critical questioning

The emphasis is on learning how to think, not memorizing recipes. AI and LLM engineering demands humility, rigor, and clarity.

This course is an invitation to:

• slow down in a fast-moving field,
• build correct mental models,
• and engineer AI systems responsibly.

If you are looking for depth over hype , this course is for you.