Varun Vemulapalli

AI Training · free course by Varun Vemulapalli

Learn AI in 100 Days.

AI explained one idea a day, in plain English, with real examples from banking, insurance, healthcare and retail. No PhD required.

By Day 100 you'll be building your own AI agents with Claude, Microsoft Copilot Studio, OpenAI Codex and GitHub Copilot.

Start with Day 1 Latest: Day 100, Reflection: 100 days later, and what's next Claude certification prep + practice exams

Phase 1 Foundations

What AI is, how machines learn, data, training vs inference, and how to judge a model. Days 1–8.

  1. 2:10 Day 1

    What is AI?

    AI is software that learns patterns from examples so it can recognize, predict or decide on situations it hasn…
  2. 2:11 Day 2

    AI vs machine learning vs deep learning vs generative AI

    AI is the umbrella, Machine Learning learns from examples, Deep Learning does it with many-layered neural netw…
  3. 2:06 Day 3

    How machines learn from examples

    A machine "learns" by guessing, measuring how wrong it was on examples with known answers, and adjusting, over…
  4. 1:41 Day 4

    Data, features and labels

    Data is the past examples, features are the clues, labels are the answers, and a model is only as good as all …
  5. 1:45 Day 5

    Training vs inference

    Training is the slow, expensive learning done ahead of time; inference is the fast answer the frozen model giv…
  6. 2:08 Day 6

    Supervised, unsupervised and reinforcement learning

    Supervised learns from answers, unsupervised finds patterns without answers, and reinforcement learns from rew…
  7. 1:53 Day 7

    Accuracy, precision and recall

    Accuracy says how often a model is right overall; precision says how much to trust its alarms; recall says how…
  8. 1:48 Day 8

    Overfitting

    Overfitting is when a model memorizes its training data instead of learning the pattern, so it looks great in …

Phase 2 Neural networks

Neurons, how networks learn, embeddings, computer vision and transfer learning, then a recap quiz. Days 9–14.

  1. 1:48 Day 9

    Neurons and neural networks

    A neuron multiplies inputs by learned weights, adds them up and decides; stack enough of them in layers and yo…
  2. 1:45 Day 10

    Loss, gradient descent and backpropagation

    A network learns by measuring how wrong it is (loss), working out each weights share of the blame (backpropaga…
  3. 1:52 Day 11

    Embeddings

    An embedding turns meaning into a list of numbers, so similar things land close together and AI can measure si…
  4. 2:00 Day 12

    How AI sees images

    AI sees an image as a grid of numbers and builds up from edges to shapes to objects, layer by layer, using fil…
  5. 1:51 Day 13

    Transfer learning and GPUs

    Transfer learning reuses what a big model already learned so you need far less data, and GPUs make the heavy n…
  6. 1:56 Day 14

    Recap and quiz: Days 1–13

    Two weeks in. If you've followed along, you now understand more about how AI works than many people who use it…

Phase 3 Large language models

Generative AI, tokens, transformers, how LLMs are trained, and how to prompt them well at work. Days 15–28.

  1. 1:47 Day 15

    What is generative AI?

    Classic AI picks from options it was given, while generative AI creates new content by predicting the next wor…
  2. 1:56 Day 16

    Tokens

    LLMs read text as tokens (about ¾ of a word each in English), and tokens drive both your bill and your limits.
  3. 1:40 Day 17

    Transformers and attention

    Transformers use attention so every word can look at every other word at once, which is how AI understands wor…
  4. 1:49 Day 18

    How LLMs are pretrained

    Pretraining teaches a model language, facts and patterns through billions of next-word guesses, but it stops a…
  5. 1:52 Day 19

    Fine-tuning and RLHF

    Fine-tuning teaches a model your format from approved examples, and RLHF teaches it what people prefer by lear…
  6. 1:51 Day 20

    Context windows

    The context window is the model's working memory in tokens; when you overfill it, things get cut or ignored, s…
  7. 1:52 Day 21

    Temperature and sampling

    Temperature decides how adventurous the model is when picking each next word: turn it up for ideas, down for a…
  8. 1:58 Day 22

    Hallucinations

    LLMs predict plausible text rather than look things up, so give them the source, ask them to cite it, let them…
  9. 1:27 Day 23

    Prompt engineering basics

    Give the model a role, a clear task, the context, the format you want and the rules to follow, and a vague ans…
  10. 1:54 Day 24

    Few-shot prompting and step-by-step reasoning

    Show the model a few labeled examples to get consistent answers in your format, and ask it to reason step by s…
  11. 1:36 Day 25

    System prompts and roles

    The system prompt sets who the assistant is and the rules it follows, the user message is the request, and the…
  12. 1:38 Day 26

    Structured outputs: JSON

    Define a schema, ask the model for JSON that matches it, and validate every result, so AI output can flow into…
  13. 1:49 Day 27

    Multimodal AI

    Multimodal AI reads images, audio and text together, which makes intake faster and richer, but a human still c…
  14. 1:49 Day 28

    Recap of Days 15–27

    An LLM predicts the next token from patterns; clear prompts, real sources and structured output make it reliab…

Phase 4 Giving AI knowledge

Semantic search, vector databases, retrieval-augmented generation, chunking, and choosing a model. Days 29–34.

  1. 1:35 Day 29

    Semantic search

    Semantic search compares meanings as embeddings, so it finds the right answer even when the user's words don't…
  2. 1:42 Day 30

    Vector databases

    A vector database stores embeddings with their metadata and quickly finds the closest matches by meaning, whic…
  3. 2:14 Day 31

    Retrieval-augmented generation (RAG)

    RAG retrieves the right passages from your own documents and gives them to the model, so answers are grounded,…
  4. 1:58 Day 32

    Chunking and retrieval quality

    Good RAG starts with good chunks: split by structure, keep tables whole, add overlap and metadata, and test re…
  5. 1:48 Day 33

    Open vs closed models, and how to choose one

    Open-weight models give you control but you run them; closed models give you ease but you trust a provider, so…
  6. 1:23 Day 34

    Recap of Days 29–33

    Phase 4 done. You now know how to give an AI model knowledge it was never trained on, and how to pick the mode…

Phase 5 How AI agents work

The agent loop, tools, memory, planning, MCP, multi-agent systems, human-in-the-loop and guardrails. Days 35–46.

  1. 1:54 Day 35

    Chatbot vs workflow vs agent

    Use a chatbot to answer, a workflow when the steps are known, and an agent only when the steps depend on what …
  2. 1:45 Day 36

    The agent loop

    An agent is a loop of think, act, observe, repeat, and the stopping conditions you give it matter as much as t…
  3. 1:51 Day 37

    Tool use / function calling

    With tool use, the model decides which function to call and with what inputs, and your app runs it and returns…
  4. 1:48 Day 38

    Designing good tools for an agent

    Give an agent small, clearly named tools with typed inputs, helpful errors and only the access each job needs,…
  5. 1:53 Day 39

    Memory and state

    Models don't remember; apps give agents memory by saving the right facts and task state and feeding them back …
  6. 1:40 Day 40

    Planning and multi-step tasks

    Good agents break a goal into ordered steps, respect what depends on what, and re-plan out loud when something…
  7. 1:58 Day 41

    MCP (Model Context Protocol)

    MCP is a shared standard that lets any compatible AI app plug into tools and data through servers, so you buil…
  8. 2:06 Day 42

    Agent patterns

    Chaining, routing, parallel and orchestrator–workers are building blocks; pick the simplest one that fits the …
  9. 2:02 Day 43

    Multi-agent systems and subagents

    Multi-agent systems give each part of a big job its own clean context, which helps on wide research but adds c…
  10. 1:58 Day 44

    Human-in-the-loop

    Human-in-the-loop lets agents handle routine cases on their own and pause for a person on high-stakes ones, wi…
  11. 1:53 Day 45

    Guardrails and prompt injection

    Anything an AI reads can try to give it orders, so treat content as data, check inputs and outputs, limit what…
  12. 1:36 Day 46

    Recap of Days 35–45 + which tool for which job

    Agents think, act and observe with tools, under guardrails and human checkpoints. From tomorrow, we build.

Phase 6 Hands-on foundations

Your setup, Git, Python, APIs, keeping keys safe, reading docs and your first call to a language model. Days 47–54.

  1. 1:49 Day 47

    Your setup: VS Code and the terminal

    The terminal is a precise text conversation with your computer, and five safe commands are enough to find your…
  2. 1:50 Day 48

    Git and GitHub basics

    Commits are save points, branches are safe copies, and pull requests are reviewed merges, so a team (or an AI …
  3. 1:41 Day 49

    Python essentials for building agents

    Variables, lists, dicts, loops, functions and packages are enough to read data and build the small tools agent…
  4. 1:48 Day 50

    APIs and JSON explained

    An API is a waiter between programs: you send a request to a URL with a method, and get back a status code and…
  5. 2:05 Day 51

    API keys, environment variables and keeping secrets safe

    Keep keys in environment variables, keep .env out of Git, and if a key leaks, revoke it first and clean up sec…
  6. 1:45 Day 52

    How to read developer docs quickly

    Read docs on a fixed path (quickstart, auth, one endpoint, errors, rate limits) and skip the rest until you ne…
  7. 1:49 Day 53

    Your first API call to an LLM

    An LLM API call is a POST with a model and a prompt; ask for JSON and the answer can feed your code.
  8. 1:54 Day 54

    Recap of Days 47–53

    Terminal, Git, Python, APIs, secrets, docs and one LLM call make up the builder's loop that coding agents will…

Phase 7 GitHub Copilot

Completions, chat, agent mode, the coding agent, custom instructions and MCP, then a hands-on build. Days 55–60.

  1. 1:60 Day 55

    What GitHub Copilot is, plans and setup in VS Code

    GitHub Copilot is an AI assistant in your editor, and the free plan plus VS Code is all you need to start.
  2. 1:46 Day 56

    Code completions and Copilot Chat

    Completions finish the line you're writing; Chat explains, questions and tests it, and you check both.
  3. 1:46 Day 57

    Agent mode in VS Code: Copilot editing across files

    In Agent, Copilot plans and edits across files, and your job becomes clear criteria and careful review.
  4. 1:47 Day 58

    Copilot coding agent: assign an issue, get a pull request

    Assign a clear issue to Copilot cloud agent and it returns a pull request, but a human still reviews and merge…
  5. 1:43 Day 59

    Custom instructions and MCP servers in Copilot

    An instructions file tells Copilot your project's rules, and MCP servers give it new tools.
  6. 1:56 Day 60

    Hands-on: build a small working tool with GitHub Copilot

    Clear rules, worked numbers and tests turn Copilot into a reliable builder, and you into the checker.

Phase 8 OpenAI Codex

The CLI, IDE and cloud agent, AGENTS.md, tests and code review, then a feature built end to end. Days 61–65.

  1. 2:00 Day 61

    What Codex is: CLI, IDE and cloud agent

    Codex is one coding agent you can use from the terminal, your editor, the ChatGPT app or the cloud, and a safe…
  2. 1:46 Day 62

    Setting up Codex and your first task

    Install Codex, sign in with ChatGPT, commit a checkpoint, and start with one failing test you can verify yours…
  3. 1:50 Day 63

    Guiding Codex with AGENTS.md

    AGENTS.md is the onboarding note for your coding agent: short, specific rules in the repo, loaded before every…
  4. 1:35 Day 64

    Codex for tests, refactoring and code review

    Have Codex write tests first, refactor second and review third, and keep a human as the final approver.
  5. 1:49 Day 65

    Hands-on: have Codex build and test a feature

    You write the spec and the tricky example; Codex plans, builds, tests and reviews; you approve the commit.

Phase 9 Claude for builders

Claude.ai, the API and tool use, Claude Code, MCP, the Agent SDK, subagents and hooks, and a comparison of the coding agents. Days 66–76.

  1. 1:46 Day 66

    Claude for builders: models, Projects and artifacts

    Pick a model for the job, put the context in a Project once, and let artifacts turn chats into things you can …
  2. 1:43 Day 67

    Your first Claude API call

    A Claude API call is a model, a length limit and your messages, sent with a secret key, turning raw text into …
  3. 1:43 Day 68

    Tool use with the Claude API

    With tool use, Claude asks for a function call in JSON, your code runs it and returns a matching tool_result, …
  4. 1:51 Day 69

    Claude Code: setup and your first session

    Claude Code is an agent in your terminal or editor that reads, runs and edits your project; give it clear expe…
  5. 1:39 Day 70

    Steering Claude Code: CLAUDE.md, slash commands and skills

    CLAUDE.md holds what Claude should always know, slash commands steer a session, and skills package a repeatabl…
  6. 1:44 Day 71

    Connecting Claude to your tools with MCP servers

    MCP servers give Claude a menu of tools for your systems, added with one command in Claude Code or as a connec…
  7. 1:46 Day 72

    The Claude Agent SDK: what it is and when to use it

    Use the API for single calls, Claude Code for hands-on work, and the Agent SDK when your own program needs an …
  8. 1:28 Day 73

    Build a research agent with the Claude Agent SDK

    A research agent is the Agent SDK loop with search, read and write tools plus clear rules, and its brief is on…
  9. 1:39 Day 74

    Subagents, hooks and permissions

    Subagents split the work, hooks make steps automatic, and permission rules set hard limits that the model can'…
  10. 1:34 Day 75

    Hands-on: Claude Code builds an agent end to end

    Describe the agent, review the plan, and let Claude Code build, test and run it, while you keep the approvals …
  11. 1:39 Day 76

    Recap of Days 55–75: comparing the coding agents

    Copilot, Codex and Claude Code share the same core skills; pick by where your work happens and keep a human re…

Phase 10 Microsoft Copilot Studio

Low-code agents for Microsoft 365: knowledge, topics, actions, autonomous triggers, publishing, governance, and choosing the right tool. Days 77–85.

  1. 1:47 Day 77

    What Copilot Studio is: low-code agents for Microsoft 365

    Copilot Studio lets you assemble an agent from instructions, knowledge, tools, topics and triggers, and publis…
  2. 1:38 Day 78

    Your first agent: instructions and knowledge sources

    Clear instructions plus well-described knowledge turn a long manual into an agent that answers with citations.
  3. 1:41 Day 79

    Topics, triggers and conversation design

    Topics give high-stakes conversations fixed, tested steps, while generative orchestration decides when to use …
  4. 1:49 Day 80

    Actions: connectors, Power Automate flows and APIs

    A tool, like an agent flow with clear inputs and outputs, turns your agent from an answerer into a doer.
  5. 1:50 Day 81

    Autonomous agents and event triggers

    Event triggers let an agent start itself, so keep instructions precise, credentials in mind and a person on fi…
  6. 1:41 Day 82

    Publishing to Teams, Microsoft 365 Copilot and websites

    Publish, pick channels where people already work, set authentication on purpose, and roll out from pilot to or…
  7. 1:49 Day 83

    Security, governance and admin controls

    Environments, data policies, sign-in and audit logs let a company say yes to agent builders without losing con…
  8. 1:35 Day 84

    Hands-on: build an HR-policy agent in Copilot Studio

    A good policy agent answers from the documents, cites them, and hands personal cases to a human.
  9. 1:42 Day 85

    Choosing the right tool: Claude, Copilot Studio, Codex or GitHub Copilot

    Pick the tool by who builds it, where users work, what it produces and what it must connect to, not by which d…

Phase 11 Capstone: build your agent

Design, spec, the agent loop, tools, knowledge, evals, cost, security, deployment, feedback and a portfolio showcase. Days 86–96.

  1. 1:50 Day 86

    Designing your agent: use case, users, scope, success criteria

    Design an agent like a product: name the job, the users, the boundaries and the test for "done" before you wri…
  2. 1:44 Day 87

    Writing the spec and system prompt

    Write the spec for humans first, then translate it into a system prompt with a role, clear rules, tags and an …
  3. 1:26 Day 88

    Building the core agent loop

    The agent loop is plain code: call the model, run the tool it asks for, send back the result, and stop when it…
  4. 1:46 Day 89

    Adding tools and actions

    Give your agent reading tools freely, but put every action behind an approval gate your code enforces.
  5. 1:28 Day 90

    Adding knowledge (RAG and knowledge sources)

    Give your agent the team's glossary and owner directory, in the prompt while small, through retrieval when lar…
  6. 1:43 Day 91

    Testing and evals: how to know it works

    An eval is a fixed exam of inputs and golden answers that turns "it seems to work" into a score you rerun afte…
  7. 1:46 Day 92

    Cost, latency and monitoring

    Log every run to know cost, seconds and errors per run; resent input tokens are usually the biggest lever.
  8. 1:49 Day 93

    Security and responsible-AI checklist

    Treat every transcript as untrusted data, give tools the least power they need, mask PII, and keep a human app…
  9. 1:40 Day 94

    Deploying and sharing your agent

    Share your agent one rung at a time, from your laptop to a teammate to a small private web app, and hand the o…
  10. 1:28 Day 95

    Collecting feedback and iterating

    Log every approval, edit and thumbs-down, fix the biggest bucket one change at a time, and turn each failure i…
  11. 1:30 Day 96

    Showcase it: README, demo video and GitHub portfolio

    A one-line README, a 90-second demo and a pinned repo turn your agent into proof anyone can check in 2 minutes…

Phase 12 AI at work and what's next

Real use cases in regulated industries, responsible AI, your roadmap and a look back at 100 days. Days 97–100.

  1. 1:54 Day 97

    Agents at work: real use cases in banking, insurance and healthcare

    Agents at work do the legwork around a decision (gather, check, draft, log) while a person keeps the decision.
  2. 1:44 Day 98

    Responsible AI: bias, privacy and regulation

    Test outcomes by group, send only the data you need, and know which rules apply before you launch.
  3. 1:32 Day 99

    Your roadmap after 100 days

    Pick one real problem and give it 90 days: use it safely, build it, then ship it and show it.
  4. 1:35 Day 100

    Reflection: 100 days later, and what's next

    You started with "what is AI?" and finished able to design, build, test and ship an agent responsibly.