Artificial Intelligence
Advanced
★ Featured
Production RAG Systems
Move retrieval-augmented generation from prototype to dependable service.
Course Overview
Cover chunking decisions, retrieval evaluation, citation quality, and operational safeguards. Exercises turn a simple search-and-answer flow into a system that can be measured and improved.
What You Will Learn
Evaluate chunking strategies (semantic, recursive, hierarchical) for enterprise documents
Implement hybrid search combining dense vector retrieval and sparse keyword indexing
Build deterministic reranking and hallucination suppression pipelines
Configure end-to-end evaluation metrics including context recall and grounding precision
Mitigate vector store data drift with transactional vector outbox patterns
Tools & Technologies Used
Qdrant
Python
FastAPI
Tiktoken
Ragas Evaluation Framework
Structured Curriculum
2 Modules · 6 Lessons · 300 Minutes Total
Module 1
Module 1: Document Processing & Retrieval Foundations
3 lessons
Module 1: Document Processing & Retrieval Foundations
Optimize chunking, metadata enrichment, and indexing strategies.
-
Semantic Chunking vs. Token Sliding Windows
Analyze chunk boundary impact on query context preservation and retrieval precision.
-
Enriching Metadata for Targeted Filtering
Extract entity tags and hierarchy markers during ingestion to enable exact-match metadata filters.
-
Dense & Sparse Hybrid Retrieval Setup
Combine vector similarity scores with BM25 keyword matching for optimal recall.
Module 2
Module 2: Reranking, Grounding & Evaluation
3 lessons
Module 2: Reranking, Grounding & Evaluation
Ensure zero hallucination and high precision before answer generation.
-
Cross-Encoder Reranking Pipelines
Implement secondary cross-encoder scoring to filter out top-K irrelevant chunks.
-
Grounding Validation & Citation Verification
Build a citation check node that asserts every statement traces back to retrieved evidence.
-
Automated RAG Evaluation Suite
Set up continuous offline evaluation benchmarking Faithfulness, Answer Relevance, and Context Precision.
Practical Project & Capstone Outcome
🚀 Capstone Project
Enterprise Technical Documentation RAG Service
Construct a production RAG backend that ingests multi-format tech docs, indexes them into Qdrant with outbox transactional sync, performs hybrid search, and serves grounded answers with verifiable source citations.
Prerequisites
- Familiarity with vector embeddings and database queries
- Intermediate Python skill level
Intended Audience
- AI Platform Engineers scaling knowledge search infrastructure
- Full Stack Developers adding reliable semantic search to products
Instructor Information
L
Leena Das
Course Author & Industry Expert
Leena Das is a Senior Retrieval Systems Engineer with expertise in high-throughput vector search and domain-adapted semantic ranking.
Frequently Asked Questions
Which vector store is used during hands-on exercises?
The course uses Qdrant, but all vector storage patterns apply equally to Pgvector or Pinecone.
How do we handle large PDF or Markdown document ingestion?
Module 1 covers specialized parser libraries and recursive layout-aware chunking.