AI Concepts • AI Hackathon

🧠 Large Language Models (LLMs)

LLMs are AI models trained on massive amounts of text data that can understand and generate human-like text. They're the foundation of modern AI assistants like ChatGPT and Claude. Foundation Text Generation

Training & Architecture

LLMs use transformer architecture and are trained on billions of text examples to learn patterns in language. They predict the next word in a sequence, which allows them to generate coherent responses.

Prompting

The way you ask questions to an LLM matters! Good prompts are clear, specific, and provide context. Techniques like few-shot learning (giving examples) can dramatically improve results.

Context Window

The "memory" of an LLM during a conversation. It's measured in tokens (roughly words). Larger context windows allow the AI to remember more of your conversation and process longer documents.

📚 RAG (Retrieval-Augmented Generation)

RAG combines LLMs with external knowledge retrieval. Instead of relying only on training data, the AI searches a database for relevant information before generating a response. Knowledge Accuracy

Why RAG Matters

RAG solves the "hallucination" problem where LLMs make things up. By retrieving real documents first, the AI can cite sources and provide accurate, up-to-date information not in its training data.

Vector Databases

RAG systems store information as mathematical "embeddings" in vector databases. This allows semantic search—finding information based on meaning, not just keywords.

Common Use Cases

Perfect for company knowledge bases, legal document analysis, customer support with product manuals, and any scenario where you need AI with access to specific, current information.

🔌 MCP (Model Context Protocol)

MCP is a standard way for AI models to connect with external tools and data sources. Think of it as USB ports for AI—a universal way to plug in capabilities. Integration Anthropic

The Problem MCP Solves

Before MCP, each AI tool integration was custom-built. MCP creates a standard protocol so developers can write one integration that works across multiple AI systems.

MCP Servers

MCP servers expose capabilities (like searching files, accessing databases, or calling APIs) that AI clients can use. Anyone can build an MCP server for their tools.

Real-World Examples

Connect Claude to Google Drive, Slack, GitHub, or your company's internal systems through MCP. The AI can then read, search, and interact with these tools directly.

🤖 AI Agents

AI Agents are autonomous systems that can plan, execute tasks, use tools, and work toward goals with minimal human intervention. They're AI that can "do things" not just answer questions. Automation Action

Agent vs. Chatbot

A chatbot responds to questions. An agent can break down complex tasks, use multiple tools in sequence, learn from feedback, and complete objectives independently.

Tool Use & Function Calling

Agents can decide when and how to use external tools—search engines, calculators, APIs, databases. They choose the right tool for each step of a task.

Multi-Step Reasoning

Agents can plan multi-step workflows: research a topic, synthesize findings, create a document, send it via email—all from a single instruction.

Examples

Research agents that compile reports, coding agents that build software, data analysis agents that process spreadsheets, customer service agents that resolve issues.

✨ Vibe Coding

A new programming paradigm where you describe what you want in natural language and AI writes the code. Less syntax, more creativity. You "vibe" the solution into existence. No-Code Natural Language

The Philosophy

Instead of memorizing syntax and debugging semicolons, you focus on the creative vision. Describe your app's behavior, design, and features in plain English.

Iterative Development

Start with a rough idea, let AI generate code, test it, then refine with more natural language instructions. Build prototypes in minutes, not days.

Tools & Platforms

Claude Artifacts, ChatGPT Code Interpreter, Replit AI, Cursor, and GitHub Copilot enable vibe coding. Each conversation becomes a collaborative coding session.

Limitations

Best for prototypes and moderate-complexity apps. Large-scale systems still need traditional engineering. The "vibe" works better when you understand what you're asking for.

🎯 Embeddings

Embeddings convert text, images, or other data into numerical vectors that capture meaning. Similar concepts have similar embeddings, enabling semantic search and comparison. Vector Math Semantic Search

How They Work

An embedding model turns "king" into something like [0.2, 0.8, -0.3, ...] with hundreds of dimensions. Words with similar meanings have vectors that point in similar directions.

Semantic Similarity

Search for "happy" and find "joyful", "delighted", "cheerful"—not just exact matches. This powers smart search, recommendation systems, and document clustering.

Applications

Powers RAG systems, semantic search engines, content recommendations, duplicate detection, and classification tasks. The backbone of modern AI information retrieval.

🎓 Fine-Tuning

Training a pre-trained model on your specific data to customize its behavior, style, or domain expertise. Like teaching an expert new specialized knowledge. Customization Specialization

When to Fine-Tune

Use fine-tuning when you need consistent style, domain-specific language, or behavior that prompting alone can't achieve. Medical diagnoses, legal writing, brand voice.

Fine-Tuning vs. RAG

RAG adds knowledge, fine-tuning changes behavior. Use RAG for facts that change. Use fine-tuning for consistent tone, format, or reasoning patterns.

Process

Collect examples of desired behavior, format as training pairs (input → desired output), run training, evaluate results. Requires hundreds to thousands of examples.

💬 Prompt Engineering

The art and science of crafting inputs that get the best results from AI models. Good prompts are clear, specific, and structured to guide the AI's reasoning process. Best Practices Optimization

Chain-of-Thought

Ask the AI to "think step by step" or "explain your reasoning." This improves accuracy on complex tasks by making the model show its work.

Few-Shot Learning

Provide 2-5 examples of the task before asking the AI to perform it. Shows the model exactly what format and style you want.

System Prompts

Instructions that set the AI's role and behavior for the entire conversation. "You are an expert teacher explaining concepts to beginners..."

Structured Output

Request specific formats: JSON, markdown tables, XML. Use delimiters and clear instructions to get consistent, parseable responses.

AI Tools & Concepts Mindmap

🤖 AI Tools & Concepts

💡 How to use this mindmap