


RAG from Zero, Ep 2
BitBanter teaches AI engineering to developers who want to understand AI, not just use it. We run every claim as an experiment, with real code and real numbers, then show you what happened.
Free. Hindi videos, English notes, code in Python and JavaScript.



RAG from Zero, Ep 2
These are real DocBot experiments from the course, replayed one after another. Each bar is a number we measured, and each run links to the notes that explain it.
See the courseReal runs from DocBot
How many days do I have to ask for a refund?
cosine similarity, top 4 chunks
The old policy ranked first. Search worked. The index was wrong.
Episode 1 notesAn embedding model turned 104 of DocBot's questions and handbook paragraphs into numbers. Similar meanings landed near each other, even across Hindi and English. Pick a question and see what it lands next to.
Try a question
Nearest neighbours by cosine similarity on the full 3,072-number vectors, not the 2D picture.
Our first course, in progress
We build one support bot, DocBot, and fix one real failure per episode. Point at an episode to see which part of the bot it covers.
We run the experiment before we explain it. If the result disagrees with the hype, we say so.
APIs, tokens, cost and code, explained so a beginner follows and a senior learns something.
Videos move fast. The notes keep the code in Python and JavaScript, the numbers and the sources.
BitBanter teaches AI engineering to developers. Each video runs a real experiment with code and numbers, then explains what happened. Videos are on YouTube and Instagram; notes and code are on this site.
Yes. The videos, the course notes and the code are all free. There is no sign-up.
The videos are in Hindi, mixed with English for technical terms. The notes on this site are in English, with code in Python and JavaScript.
Start with RAG from Zero, a free course where we build one support bot, DocBot, and fix one real failure per episode. Episode 1 explains what RAG is; Episode 2 explains embeddings.
YouTube (@bitbanteryt) has the full breakdowns and course episodes, usually 5 to 7 minutes. Instagram (@bitbanter.ig) has short reels on AI tools, tests and tech news.
Developers and computer science students who use AI tools and want to understand how they work: embeddings, RAG, agents, tokens and cost.

Short reels on Instagram, full breakdowns on YouTube.