Report a bug

What went wrong?

RAG & Vector Databases — AI Academy

AI Academy/Courses/RAG & Vector Databases

⌘K

UK Sign in Sign up

IntermediateИнженерия

RAG & Vector Databases

Complete guide to Retrieval-Augmented Generation: from embeddings and chunking to production RAG pipelines with Pinecone, Qdrant, pgvector. Covers semantic search, hybrid search (BM25 + dense), reranking, evaluation (RAGAS), and patterns for documents, code, and multimodal data.

4modules

12lessons

360 mintotal time

Developers building search and Q&A systems on LLMsaudience

Progress

0%0 / 12 lessons

Sign in to save your progress.

Module 1

Embeddings & Vector Search

What embeddings are, how similarity is computed, which models to choose, and where to store vectors.

Embeddings and Similarity Metrics

What an embedding is, cosine vs dot product vs Euclidean, comparing text-embedding-3-large, ada-002, Cohere Embed v3 and all-MiniLM, choosing dimensions.

Vector Stores: Pinecone, Qdrant, pgvector, Chroma, FAISS

Comparing managed and self-hosted stores, HNSW and IVF index types, running Qdrant in Docker and pgvector in Postgres, when to choose what.

Full Semantic Search: embed → upsert → query

End-to-end search pipeline: embedding, upsert with metadata, top-k query, metadata filtering, and namespace partitioning.

Module 2

Chunking & Preprocessing

How to split documents, load different formats, and clean data before indexing.

Chunking Strategies

Fixed-size, sentence, semantic chunking, overlap, the impact of chunk size on recall vs precision, and the small-to-big retrieval pattern.

Document Loaders: PDF, DOCX, HTML, Markdown, code

Extracting clean text from PDF (PyMuPDF, pdfplumber), DOCX, HTML (BeautifulSoup), Markdown and code while preserving structure.

Preprocessing: deduplication, cleaning, metadata

Removing duplicates, noise and non-informative chunks, extracting metadata (page, section, source URL) before indexing.

Module 3

Advanced Retrieval

Hybrid search, reranking, and query expansion for complex questions.

Hybrid Search: BM25 + dense, RRF

Combining lexical BM25 and dense vectors, reciprocal rank fusion (RRF), Weaviate hybrid, why hybrid beats pure semantic on keyword-heavy queries.

Reranking: cross-encoder and two-stage retrieval

Cross-encoder rerankers (Cohere Rerank, BGE reranker), the recall→precision two-stage scheme, latency and cost trade-off.

Query Expansion: HyDE, multi-query, step-back

HyDE (hypothetical document embedding), multi-query generation, step-back prompting and query decomposition for complex questions.

Module 4

RAG Pipeline & Evaluation

An end-to-end pipeline, automated evaluation via RAGAS, and production patterns.

End-to-End RAG Pipeline

Ingest pipeline (embed + store) and query pipeline (retrieve → rerank → prompt → generate), comparing LangChain, LlamaIndex and the raw SDK.

RAG Evaluation: RAGAS

The RAGAS framework: context precision, context recall, faithfulness, answer relevancy — automated evals and a regression test suite.

Production RAG: streaming, citations, cache, monitoring

Streaming responses, citation generation tied to chunks, caching embeddings, monitoring retrieval quality, and cost optimization.

AAI AcademyEN Start learningCatalog

Overview

AI Academy Progress Achievements Leaderboard Insights Notes Notifications Team API Pricing Settings

Learning

Tracks8 Courses51

Libraries

Prompts30 Workflows12 Tools22 Compare Blog

Updated · May 14, 2026