LLM Interview Questions

🡡️ Open the latest version (GitHub Pages)

An interactive learning experience with hands-on examples for the most important Large Language Model (LLM) interview questions.

🏯 Overview

This project is a single-page, static web app covering an expanding set of essential LLM questions. Every question includes a clear explanation and most include interactive mini-simulators. There's also a searchable Glossary with deep links into the questions.

🎯 Features

• Curated questions with explanations and interactive widgets (continuously growing)
• Searchable in-app Glossary with "Learn more: Question N →" cross-links
• Shareable deep links per question (Share button or press "S")
• Keyboard navigation (← / →), and a top dropdown to jump anywhere
• Math rendering via MathJax (inline and display equations)
• Lightweight: pure HTML/JS/CSS; no build step needed; mobile-friendly

🤖 Interactive Tutorials

In addition to the Q&A format, this repository includes comprehensive interactive tutorials:

Agentic Search Tutorial

Launch Tutorial →

An interactive deep-dive into agentic search, teaching marketers and developers how AI agents discover and use tools through the Model Context Protocol (MCP).

What you'll learn:

• Traditional search vs. agentic search comparison
• How AI agents discover tools through .well-known/ai-plugin.json manifests
• The Model Context Protocol (MCP) and tool discovery
• 6 realistic customer-facing scenarios (bike shop example)
• FEED optimization framework with real conversion data (15.9% vs 1.8%)
• Common limitations and how to address them

Features:

• 6 interactive sections with hands-on simulations
• Toolbox explorer with 8 interactive tools
• Live scenario player showing step-by-step agent reasoning
• Conversion metrics and optimization strategies
• Dark/light theme support
• Mobile-responsive design

Tech stack: React + TypeScript + Vite + Framer Motion

Location: /tutorials/agentic-search/

📡 Topics covered (selection)

• Tokenization, embeddings, attention and multi-head attention
• Context window, KV cache, decoding (greedy, beam, top-k/top-p, temperature)
• LoRA/QLoRA and PEFT, RLHF and alignment
• MoE, RAG, few/zero-shot prompting, CoT
• Math foundations: softmax, cross-entropy, gradients, Jacobians, eigenvalues
• Deployment trade-offs: latency, cost, safety, privacy, quantization, distillation

🛰️ Technologies

• HTML + Tailwind CSS (CDN)
• Vanilla JavaScript
• MathJax v3 (tex-svg)

📁 Project structure

top-50-llm-questions/
|-- index.html                 # App shell (header, viewer, glossary modal, footer)
|-- js/
|   |-- app.js                # SPA navigation, deep links, caching, MathJax hooks
|   |-- questionLoader.js     # Manifest-driven loader for question assets
|   \-- glossary.js           # Searchable glossary with in-app navigation
|-- questions/
|   |-- manifest.json         # Maps ids to ./questions/qXX folders
|   |-- q01/ ... q57/         # Question assets (answer/interactive files)
|   \-- q-template/           # Copy to qXX/ when authoring a new question
|-- papers/
|   |-- manifest.json         # Maps paper ids to ./papers/pXX folders
|   |-- p-template/           # Copy to pXX/ when authoring a new paper
|   \-- p01/                  # Paper assets (overview/interactive files)
|-- tutorials/
|   |-- index.html            # Tutorials landing page
|   \-- agentic-search/       # Interactive agentic search tutorial (React + Vite)
|       |-- src/              # React components and TypeScript source
|       |-- dist/             # Production build (deployed to GitHub Pages)
|       \-- README.md         # Tutorial-specific documentation
|-- QUESTION_TEMPLATE_GUIDE.md            # Question authoring guidance
|-- QUESTION_CHECKLIST.md                 # Question review & test checklist
|-- PAPER_TEMPLATE_GUIDE.md               # Paper authoring guidance
|-- PAPER_CHECKLIST.md                    # Paper review & test checklist
|-- COPILOT_SYSTEM_PROMPT.md              # AI-assisted coding system prompt
|-- LICENSE                               # MIT License
\-- README.md

🏮 How to run

1. Download or clone the repo
2. Open index.html in a modern browser (no server needed)

Tips inside the app:

• Use the dropdown or ← / → to navigate
• Press "S" to copy a shareable deep link to the current question
• Click "Glossary" in the header to browse definitions and jump to questions

🏧 Development

Tooling

Install the lint tooling once:

npm install

Run the repo checks before shipping:

• npm run lint:html — catches Tailwind colour utilities or inline colours in HTML fragments.
• npm run lint:css — blocks raw colour literals outside the shared theme overrides.
• npm run lint:repo — validates manifest links and interactive scaffolding.

Adding a new question

1. Copy questions/q-template/ to questions/qXX/ (two-digit id).
2. Customize answer.html, interactive.html, and interactive.js (ensure it exports interactiveScript).
3. Add the id to questions/manifest.json with a dir entry and update interactiveTitle.
4. Verify availableQuestions (and learning paths) in js/app.js, update /q/XX.html, then follow QUESTION_TEMPLATE_GUIDE.md and QUESTION_CHECKLIST.md.

Question assets:

• answer.html: HTML fragment rendered inside the viewer
• interactive.html: markup for controls/results
• interactive.js: exports interactiveScript (CommonJS + browser global)
• MathJax: call window.MathJax?.typesetPromise after injecting new math markup

Adding a new paper explainer

1. Copy papers/p-template/ to papers/pXX/ (two-digit id).
2. Populate overview.html (executive quick take, callouts, evidence, roadmap) and interactive.html/interactive.js (embedding controls, coverage outlook, LIMIT-style simulator).
3. Add the entry to papers/manifest.json with dir, author metadata, summary, and relatedQuestions; update p/XX.html based on p/_template.html.
4. Hard-refresh index.html#paper-XX to verify the landing card, interactive, and related-question links. Follow PAPER_TEMPLATE_GUIDE.md and PAPER_CHECKLIST.md for detailed steps.

Paper assets:

• overview.html: HTML fragment rendered inside the paper viewer
• interactive.html: markup for the stress tester controls/results
• interactive.js: exports interactiveScript (CommonJS + browser global)
• Scenario data: define realistic queries, expected documents, and hints inside the simulator configuration

Notable implementation details

• Custom loader reads questions/manifest.json and fetches HTML/JS assets per question directory
• Adjacent questions are opportunistically preloaded to reduce perceived latency
• MathJax rendering is retried on transient errors; see index.html startup config

AI-assisted coding prompt

See COPILOT_SYSTEM_PROMPT.md for repository-wide rules when using AI assistants.

• Commit/push only on explicit user request
• Assume a local test server at http://127.0.0.1:5501 for smoke checks

📽 Attribution

Questions are based on the overview here: Top 50 LinkedIn LLM interview questions (LinkedIn): https://www.linkedin.com/posts/hoang-van-hao_top-50-linkedin-llm-interview-questions-activity-7332959385280778240-lyU0/

🏤 Contributing

Contributions are welcome. Please:

1. Fork the repo and create a feature branch
2. Follow the question contract and style conventions
3. Test thoroughly (try multiple browsers)
4. Open a PR with a concise description and screenshots/GIFs if UI changes

📔 License

This project is open source under the MIT License.

🏾 Recent updates

• NEW: Interactive Agentic Search Tutorial - comprehensive guide to AI agents and MCP
• Added question 51 (end-to-end LLM lifecycle) and interactive pipeline explorer
• Added questions 46-50 with interactive explorers
• Added searchable Glossary with cross-links to questions
• Improved deployment trade-offs explorer (Q50) and math rendering resilience
• Footer attribution and MIT LICENSE added

---

Built for the LLM community.