Kishor.Academy
Log in
← All courses

Artificial Intelligence · beginner

Large Language Models: From Prompts to Practice

Understand tokens and context, build reusable prompts, work with source documents and JSON, and evaluate a prompt portfolio honestly.

With Admin

beginner9 lessons · 3 sections6–9 hours of reading and practice, plus 2–3 hours for the project; estimates exclude video.

THE TAKEAWAYS

What you’ll learn

  • Explain tokens, context, next-token generation, and why word count is not token count.
  • Describe attention and transformers conceptually without treating attention as factual verification.
  • Distinguish pre-training, fine-tuning, alignment, and inference.
  • Compare hosted, open-weight, and local deployment tradeoffs without assuming all are free.
  • Use zero-shot and few-shot templates with explicit inputs, labels, and constraints.
  • Distinguish reasoning prompts, retrieval, and tools; validate their outputs.
  • Recognize hallucinations, knowledge gaps, and unsupported confidence.
  • Define prompt templates and score output structure separately from factual correctness.
  • Build a tested prompt portfolio with at least six cases and a revision log.

Is this for you?

Everyday AI users, educators, and future application builders who want reliable language-model workflows before coding an application.

Before you start

  • Complete AI for Everyone or know how to write and check a basic prompt.
  • A text editor and fictional source notes; no required model subscription.
  • Python 3.10 or newer only for the optional response-evaluation lab.

PUT IT TO WORK

What you’ll build

Tested prompt portfolio. Client scenario: an instructor wants help with announcements, event-fact extraction, and revision questions. Deliverable: three templates, six or more test cases, a scoring rubric, results, and a revision log.

Explore what a language model receives and generates, why fluent answers can be unsupported, and how prompting, retrieval, and fine-tuning differ. Build templates for extraction, rewriting, classification, and grounded questions. Test complete, missing, contradictory, and malformed inputs rather than selecting only successful outputs. The optional offline evaluation script scores fixed teaching examples; it is not a benchmark of real models.

Course curriculum

SECTION 1Language models, context, and prompt design3 lessons
  1. Tokens, context windows, and next-token predictionReading
  2. Transformers and attention, conceptuallyReading
  3. Pre-training, fine-tuning, and alignmentReading
SECTION 2Structured tasks and source-based answers3 lessons
  1. Hosted, open-weight, and local modelsReading
  2. Zero-shot, few-shot, and role promptingReading
  3. Reasoning, retrieval, and tool useReading
SECTION 3Evaluation, limitations, and a prompt portfolio3 lessons
  1. Hallucinations, knowledge gaps, and overconfidenceReading
  2. Prompt templates and output rubricsReading
  3. Capstone: a prompt pack for a real taskReading