How this Generative AI track works
- Read-focused — no in-browser LLM lab —
execution_profile: none. Learn transformers, prompting, RAG, and safety; practice SDK patterns locally with your own API keys. - Prerequisites — AI literacy, Python, and Data Science workflow; NumPy helps with embeddings.
- You will ship smarter — grounded assistants with citations, guardrails, eval sets, and cost-aware model choice—not hype demos.
- Pair with — DSA for retrieval complexity, AWS for Bedrock hosting, Cybersecurity for injection and abuse threats.
Never paste production API keys or customer transcripts into lessons. Use synthetic eval data in sandboxes.
Install on your device (macOS, Linux, Windows)
Read-focused track—use any modern browser and an editor for notes and diagrams.
macOS
- Browser: Safari or Chrome. Editor: VS Code (
brew install --cask visual-studio-code).
Linux
- Browser: Firefox/Chromium. Editor: VS Code or distro package manager.
Windows
- Browser: Edge/Chrome. Editor: VS Code (
winget install Microsoft.VisualStudioCode).
Verify: You can take notes and complete MCQs on this site.
Optional API practice: set provider keys in .env on your own machine—never commit secrets.
Generative AI creates new text, images, audio, or code from learned patterns—most visibly through large language models (LLMs) and multimodal assistants. This track teaches how they work, how to prompt them, how to ground answers with retrieval, and how to ship features responsibly after AI literacy, Python, and Data Science.
Prerequisites
Complete AI for ML/DL vocabulary, ethics, and product context. Comfort with Python, JSON, and HTTP helps you follow SDK examples. NumPy intuition helps with embeddings; DSA helps with retrieval complexity.
We do not train billion-parameter models here—you will design, evaluate, and integrate systems that call foundation models via APIs or open weights.
What you will learn
- Transformers, tokens, context windows, and decoding at a builder-friendly depth
- Prompt engineering: roles, few-shot, chain-of-thought, structured outputs
- RAG: chunking, vector search, citations, and failure modes
- Safety: hallucinations, PII, guardrails, human review
- Integration: Python SDK patterns, streaming, cost/latency trade-offs
- Production checklist and interview framing for Gen AI roles
Sanity check (local, optional)
# Optional — no API key required
print("Generative AI track — read-focused")
stack = ["prompt", "retrieve", "generate", "evaluate", "guard"]
print("Builder loop:", " → ".join(stack))Practice: Copy optional Python snippets into a local notebook. This track is read-focused (execution_profile: none)—learn prompting, RAG, and safety without API keys in the browser.
API keys and data handling
Never paste production API keys, customer transcripts, or regulated PII into lessons or public repos. Use vendor sandboxes with synthetic data. Read retention and training opt-out policies before sending content to third-party APIs.
Important interview questions and answers
- Q: Why Gen AI after AI literacy?
A: Generative systems inherit data, evaluation, and fairness issues—shared vocabulary prevents treating LLMs as magic. - Q: Why no in-browser Gen AI lab?
A: API cost, key leakage, and policy variance make a read/quiz track safer until a gated lab ships.
Self-check
- Name two prerequisite tracks linked from this intro.
- List the five-step builder loop printed in the optional snippet.
Challenge
Design a tiny RAG note
- Pick one internal doc topic (handbook, FAQ).
- Sketch chunk → embed → retrieve → prompt steps on paper.
- Write one refusal rule when context is missing.
Done when: you have a one-page RAG sketch with an explicit INSUFFICIENT_CONTEXT policy.
Tip: Finish AI literacy first—Gen AI assumes you know train/test, bias, and product trade-offs.
Interview prep
- Prerequisites?
AI literacy, Python, data science basics; optional NumPy for embeddings intuition.
- Read-focused why?
execution_profile: none—no API keys or model weights in the browser; study design and safety first.