# BitBanter > Learn AI engineering with tested, simple explanations. Free courses, notes and code in Python and JavaScript, from the BitBanter videos. BitBanter teaches AI engineering to developers. Each video runs a real experiment with code and numbers, then explains what happened. Videos are in Hindi with English technical terms; notes, code (Python and JavaScript) and quizzes on this site are in English. Everything is free, with no sign-up. - YouTube: https://www.youtube.com/@bitbanteryt (full breakdowns and course episodes) - Instagram: https://www.instagram.com/bitbanter.ig/ (short reels on AI tools, tests and tech news) ## Courses - [RAG from Zero](https://bitbanter.in/courses/rag-from-zero/): Free RAG full course in Hindi, beginner to pro. Retrieval Augmented Generation, built up one piece at a time. We make one support bot, DocBot, answer questions from a company handbook, and every episode fixes something real that went wrong in the last one. - [AI Agents from Zero](https://bitbanter.in/courses/ai-agents-from-zero/): GenAI, LLMs, tools and agents, learned by playing with them. What GenAI is, what an LLM actually does, why tokens decide your bill, how a chat really works, and how tools turn a model into an agent. No videos: every lesson is short reading plus puzzles you solve in the page, and each one builds on the last until you can read an agent loop line by line. ## Course notes - [RAG from Zero, episode 1: What is RAG? Why AI never read your docs](https://bitbanter.in/resources/rag/ep1-what-is-rag/): Your company's docs were never in the model's training data. RAG fixes that without training anything: search first, then ask. We open the exact message our demo bot sends and run the same question with and without RAG. - [RAG from Zero, episode 2: Embeddings: how AI knows “refund” = “paisa wapas”](https://bitbanter.in/resources/rag/ep2-embeddings/): Text becomes numbers that capture meaning. We build the idea from a 2-number toy map to 3,072 dimensions, then test it on DocBot's real data, where one model puts a Hinglish question in the wrong place. - [AI Agents from Zero, lesson 1: What GenAI actually is](https://bitbanter.in/learn/agents/l1-what-is-genai/): Your bank's fraud check and ChatGPT are both called AI, but they do different jobs. One picks an answer from a fixed list, the other writes something that didn't exist before. Learn to tell them apart in a minute. - [AI Agents from Zero, lesson 2: What an LLM actually does](https://bitbanter.in/learn/agents/l2-what-is-an-llm/): An LLM does one thing: given some text, it guesses the next small piece, then does it again. Be the model yourself, turn the temperature dial, and see why it sometimes makes things up with total confidence. - [AI Agents from Zero, lesson 3: Tokens and the context window](https://bitbanter.in/learn/agents/l3-tokens-context-window/): Models don't read words, they read tokens, and you pay per token. Guess how real text gets split, see why Hindi used to cost three times more, and shrink a context window until the model forgets who you are. - [AI Agents from Zero, lesson 4: How a chat really works](https://bitbanter.in/learn/agents/l4-messages/): The model has no memory. Every time you hit send, your app sends the whole conversation again. Step through a real chat, learn the three message roles, and see why long chats get slower and pricier. - [AI Agents from Zero, lesson 5: Tools: letting the model look things up](https://bitbanter.in/learn/agents/l5-tools/): A model can't check your database or send a refund. But it can ask your code to. Play the model in a real tool call, see exactly which messages go back and forth, and learn why the model never runs anything itself. - [AI Agents from Zero, lesson 6: Agents: an LLM in a loop](https://bitbanter.in/learn/agents/l6-agents/): An agent is not magic. It's the tool call from lesson 5, repeated in a loop until the job is done. Build the loop, run a refund agent step by step, and catch a prompt injection hiding in customer reviews. ## Blog - [Agentic RAG vs Traditional RAG: What Changes, What It Costs](https://bitbanter.in/blog/agentic-rag-vs-rag/): Traditional RAG searches once. Agentic RAG lets the model search as often as it needs. A real retrieval test, a 40-line agent loop, and when each one wins. - [RAG Pipeline: Architecture, Steps and Real Numbers](https://bitbanter.in/blog/rag-pipeline/): The 9 steps of a RAG pipeline, from loading docs to the final answer, with a framework-free build, measured costs, and the bug each step can cause. ## Latest videos - [What are embeddings? How AI turns text into numbers](https://www.youtube.com/watch?v=aCijGda2WbY) (RAG from Zero, Ep 2, 2026-11-19): Embeddings turn text into numbers that capture meaning, so "refund" and "paisa wapas" land in the same place. Built from a 2-number toy up to a real model. - [What is RAG? Why AI doesn't know your docs](https://www.youtube.com/watch?v=5WKD1Nr7PQ8) (RAG from Zero, Ep 1, 2026-11-16): Your company's docs were never in the model's training data. RAG means search first, then ask. We run the same question with and without RAG. - [Why Microsoft 365 went down](https://www.youtube.com/watch?v=PgNOM7_XUiY) (Outage breakdown, 2026-11-17): Outlook and Copilot failed while Teams calls kept working. We read Microsoft's status page and four major cloud outage reports, then crashed a fake 10,000-server company. ## Optional - [Full text of all notes and posts](https://bitbanter.in/llms-full.txt) - [RSS feed](https://bitbanter.in/rss.xml)