Learn RAG: Build Search That Works
Search isn’t magic. It’s engineering. 80% of users won’t look past the top 5 results. If those results are bad, they leave.
RAG (Retrieval Augmented Generation) is how you build search that gives users what they need.
I just launched a RAG course on Boot.dev, here’s what you’ll learn
100% of the explanations and content in the course are free, but you can pay for the interactive & submission features after the first few chapters if you want. Use code ISAACRAG for a discount, or just take the whole thing free and @ me on socials with questions.
Why This Matters
Keyword search fails when users misspell words or use different terminology. Semantic search fails when users know exactly what they want. Keyword search dominates in technical disciplines. Semantic search wins in conceptual queries.
You need both, so we build hybrid search.
Documents can be huge and unwieldy, so we build chunking algorithms.
Users leave if results are slow. They also leave if they’re wrong. So we build fast hybrid search with a thorough re-ranking for quick retrieval and accurate results.
User inputs are unpredictable, so we add LLMs for rewriting, spell correction, and query expansion to stabilize our search.
None of this matters commercially if we can’t measure it to know if it’s good. So we build evals and metrics.
Most users don’t want a list of results—they want a summary with citations or a direct answer. We build that too.
We also cover agentic search for use cases like coding agents, and multimodal search that accepts images as queries.
Prerequisites
All you need is python programming experience to take the RAG course. But if you don’t have programming experience, you can take the python courses on boot.dev to fill that pre-requisite (all included).
What You’ll Build
You’ll build a search system from scratch. You’ll implement every piece, test it locally, and see what works.
Note: Big thanks to the boot.dev team for their extensive support and work on this course including Lane, Theo, Stass, and Alex.
You’ll work with a movie database for a streaming service called Hoopla. By the end, your system will:
Find movies using keyword search (TF-IDF, BM25)
Find movies using semantic search (embeddings, cosine similarity)
Combine both with hybrid search using reciprocal rank fusion
Fix typos, rewrite queries, and expand searches with LLMs
Re-rank results for better top-5 accuracy
Answer questions directly instead of just returning documents
Summarize results from multiple sources
Search using images to find similar movies (multimodal search)
Every lesson builds on the last. One concept at a time. Each lesson is small, focused, and executable. You’ll understand what you’re building and why.
What You’ll Learn
Keyword Search: Build an inverted index. Implement TF-IDF scoring. Understand term frequency saturation and document length normalization. Implement BM25.
Semantic Search: Generate text embeddings. Calculate dot product and cosine similarity. Implement semantic search with pre-trained models. Understand when vector databases are necessary.
Hybrid Search: Combine keyword and semantic search. Normalize scores. Implement weighted combinations and reciprocal rank fusion.
Query Enhancement: Use LLMs to fix spelling errors. Rewrite queries for better results. Expand queries to capture user intent.
Re-ranking: Implement two-stage search. Use LLM re-rankers and cross-encoders and understand the tradeoffs.
Augmented Generation: Summarize multiple documents. Answer questions directly. Resolve conflicts between sources. Add citations to generated answers.
Evaluation: Build a golden dataset. Calculate precision, recall, and F1 score. Use LLMs to evaluate results. Understand what good results look like.
Chunking: Break documents into smaller pieces. Implement semantic chunking. Handle edge cases. Understand what late chunking and ColBERT approaches are.
Agentic RAG: Recursive search uses results to guide the next search. Agentic search gives the LLM tools and lets it decide what to search next. Learn when to use each.
Multimodal Search: Generate image embeddings. Search using images instead of text. Combine text and image queries in the same vector space.
What Makes This Course Different
You’ll build and test everything locally. Every lesson has CLI tests that run locally and submit to the platform for grading. You’ll know immediately if you got it right.
You’ll build incrementally with no massive leaps. Each lesson teaches one concept and only adds one piece.
You’ll understand the tradeoffs: speed vs. accuracy, keyword vs. semantic, pre-computed vs. real-time. You´ll know what to prioritize.
You’ll see the failures. Search systems fail in interesting ways. You’ll learn what breaks and how to fix it.
What You Get
This course includes access to Boot.dev’s whole platform, including their AI Agent course, and really a solid foundation in EVERYTHING a developer would need.
Developer who wants to learn to build AI stuff? Great, there’s RAG and agent building courses.
Never coded before? Great, start with the learn python introductory course and do the full curriculum.
Already know RAG/Agents/AI and want to build a product you can deploy? Great, dive into web servers, docker, kubernetes, etc.
Search is everywhere. Every app needs it. Most apps get it wrong. Learn how to get it right.
Here’s the link to the course that applies the ISAACRAG promo code if you choose to buy after taking a look at the contents