Retrieval Augmented Generation (RAG)
by DeepLearning.AI
Design, build, and evaluate production-ready RAG systems end to end, from retrievers to observability.
Overview
Retrieval Augmented Generation (RAG) is a DeepLearning.AI course (also on Coursera, in partnership with Together AI) taught by AI engineer and educator Zain Hasan, giving practitioners the conceptual grounding and hands-on experience to design, build, and evaluate production-ready RAG systems. Structured in five modules — RAG Overview; Information Retrieval and Search Foundations; Information Retrieval with Vector Databases; LLMs and Text Generation; and RAG Systems in Production — the ~26-hour program (49 video lessons and 10 graded assignments) walks through retriever architecture, keyword search with TF-IDF and BM25, semantic search and vector embeddings, hybrid search, approximate-nearest-neighbor algorithms, vector databases via the Weaviate API, chunking, query parsing, cross-encoders and reranking, prompt engineering, hallucination handling, agentic RAG, the RAG-versus-fine-tuning tradeoff, and production concerns including evaluation, logging, monitoring, observability with Phoenix from Arize, quantization, and multimodal RAG. Labs use real-world datasets from e-commerce, media, and healthcare domains.
At a Glance
- Topic
- RAG
- Level
- Intermediate
- Format
- Course
- Cost
- Freemium
- Duration
- ~26 hours, self-paced (49 video lessons, 10 graded assignments)
- Provider
- DeepLearning.AI
- Hands-on
- Yes — code/exercises
- Certificate
- Available
What You’ll Learn
- ✓Build retrievers using keyword search (TF-IDF, BM25), semantic search with embeddings, and hybrid search
- ✓Use vector databases (via the Weaviate API), approximate-nearest-neighbor indexes, chunking, and query parsing
- ✓Improve relevance with cross-encoders and reranking, and choose the right RAG architecture per use case
- ✓Handle generation: prompt engineering, hallucination mitigation, agentic RAG, and RAG-vs-fine-tuning tradeoffs
- ✓Evaluate, log, monitor, and observe RAG systems in production using Phoenix from Arize
- ✓Ship advanced production RAG including quantization and multimodal retrieval
Highlights
- •Full production arc — from retrieval fundamentals through evaluation, observability, and deployment
- •Hands-on graded assignments on real e-commerce, media, and healthcare datasets
- •Taught by Zain Hasan with tooling like Weaviate and Arize Phoenix; certificate available
Who It’s For
Best For
- ✓AI engineers building or productionizing retrieval-augmented LLM applications
- ✓Developers who want to evaluate and monitor RAG quality, not just prototype it
- ✓Practitioners choosing between RAG and fine-tuning for grounding LLMs on their data
Prerequisites
- •Intermediate Python
- •Basic familiarity with LLMs and embeddings
- •Comfort running notebooks / coding assignments
FAQ
What is Retrieval Augmented Generation (RAG)?
A hands-on DeepLearning.AI course taught by Zain Hasan (Together AI) that takes AI engineers from retrieval fundamentals to production RAG. It covers search and vector databases, prompt and retrieval strategies, evaluation, and deployment, with graded coding assignments on real-world datasets.
Is Retrieval Augmented Generation (RAG) free?
Retrieval Augmented Generation (RAG) offers free content, with paid options for certificates or premium features.
What level is Retrieval Augmented Generation (RAG) for?
Retrieval Augmented Generation (RAG) is aimed at a intermediate audience. Recommended background: Intermediate Python, Basic familiarity with LLMs and embeddings, Comfort running notebooks / coding assignments.
How long does Retrieval Augmented Generation (RAG) take?
Expect roughly ~26 hours, self-paced (49 video lessons, 10 graded assignments). Most learners work through it at their own pace.
What will I learn from Retrieval Augmented Generation (RAG)?
You'll learn: Build retrievers using keyword search (TF-IDF, BM25), semantic search with embeddings, and hybrid search; Use vector databases (via the Weaviate API), approximate-nearest-neighbor indexes, chunking, and query parsing; Improve relevance with cross-encoders and reranking, and choose the right RAG architecture per use case; Handle generation: prompt engineering, hallucination mitigation, agentic RAG, and RAG-vs-fine-tuning tradeoffs; Evaluate, log, monitor, and observe RAG systems in production using Phoenix from Arize; Ship advanced production RAG including quantization and multimodal retrieval.