AI fundamentals for Beginners - Learn LLM, Agentic AI, MCP

Differentiate traditional AI from generative AI, noting that traditional AI does not generate new content while generative AI creates new text, images, and videos using deep learning.

Learn how large language models, trained on billions of words and billions of parameters, generate human-like text and perform tasks like answering questions, translating, and content creation.

Understand how large language models predict the next token from context using billions of parameters. They learn patterns from terabytes of data, not memorize databases, enabling reasoning and multilingual understanding.

Explain tokens as the building blocks language models use to process text, including words, parts of words, numbers, punctuation, and spaces, and how tokenization influences input and pricing.

Explore how inference drives output from trained models, how the context window limits tokens, and why hallucinations occur, with rag as retrieval augmented generation.

Learn why prompt engineering matters, choosing concise prompts to save tokens and improve accuracy, and discover how to specify output formats in prompts for large language models.

Explore how temperature, top-k, and top-p sampling shape large language model outputs, from deterministic coding tasks to creative brainstorming, by adjusting randomness and probability mass.

Master prompt design by defining a persona, supplying context and input data, detailing instructions, choosing an output format, and using examples to align llm outputs with your goals.

Explore zero shot prompting, giving the model only the task with no example and relying on pre-trained knowledge for simple, well known questions to yield quick results.

Learn few-shot and one-shot prompting, where models follow examples to control tone, format, and reasoning for structured outputs, rewrites, and data transformations, including polite customer service style and chain-of-thought prompting.

Explain chain-of-thought prompting and the react prompting technique, guiding models to think step by step to improve reasoning in math, logic, and multi-step planning, with ChatGPT examples.

Explore react prompting, a reason-and-action approach where AI agents alternate thinking and acting—reasoning steps, then retrieving facts by searching the web or calling tools and API calls to produce answers.

Master system, contextual, and role prompting to design prompts that govern AI behavior, supply background context, and define persona, shaping accurate, safe, and helpful responses in AI applications.

Explore how AI agents use LLM reasoning and tools to perceive data, break prompts into steps, and autonomously execute tasks like weather checks and flight bookings.

Discover the core components of AI agents, from the LLM brain and planning to tools, memory (short- and long-term), knowledge bases, and guardrails that enable secure, task-driven actions.

Explore the AWS Bedrock agent, a fully managed, serverless service that hosts agents without code. Attach functions via Lambda, integrate with RDS, S3, Salesforce, and use Rag for data access.

Explore the architecture of an ai ops agent built on AWS Bedrock, with two Lambda tools for CloudTrail and EC2 management, and learn the required AWS CLI and IAM permissions.

Set up AWS CLI credentials by creating an IAM user in the IAM Identity Center with MFA, configure AWS SSO, verify with sts get-caller-identity, and prepare CloudFormation deployment.

Deploy and test an AI ops agent on AWS Bedrock, wiring CloudTrail, EC2, and Lambda via CloudFormation templates, with API schemas, guardrails, and orchestration to monitor and manage EC2 instances.

Explore responsible AI principles—fairness, privacy and security, accountability, reliability, and transparency—and learn how guardrails, human in the loop oversight, and AWS Bedrock tools enable safe AI.

Understand retrieval augmented generation (rag): an ai framework that improves llm accuracy by retrieving data from external trusted sources and grounding responses using vector stores, embeddings, and semantic search.

Demonstrate a restaurant chatbot using a rag pipeline with vector search in astradb and olama embeddings, configured in langflow; compare with a non rag version to show context vs hallucination.

Build a rag-based front desk agent with Langflow, using AstraDB vector store and Olama embeddings to power retrieval augmented generation from a restaurant pdf menu.

Compare traditional RAG with agentic RAG, showing how a retrieval agent enables planning, reasoning, and tool use with LLMs across vector stores, web search, and APIs for multi-agent scenarios.

Discover how MCP, the model context protocol, unifies access to services like YouTube, AWS, and Google Calendar, letting a client talk to MCP servers and run tasks without writing code.

Explore using the Docker desktop MCP server to generate YouTube transcripts by connecting a client (cloud desktop, VS Code, Gemini CLI) to MCP servers from the catalog.

Discover the basic MCP architecture that connects ai applications to external sources via the model context protocol, including hosts, clients, servers, and data and transport layers.

Build and run your own MCP server to manage Google Calendar, including OAuth authentication, API access, and client-server communication, enabling calendar events to be listed, scheduled, and deleted.

Wraps up by revisiting open-source models you can install locally, highlights prompt engineering, agentic AI with AWS bedrock, and MCP (model context protocol) for connecting external sources like Google Calendar.