*We customize the course outline and content to your specific needs and relevant use cases.
Module 1: Designing a production oriented RAG architecture
- Separating ingestion, indexing, retrieval, augmentation, and generation into clear components
- Designing interfaces between document sources, embedding services, indexes, and language models
- Choosing between synchronous and asynchronous processing for different parts of the system
- Planning for document updates, reindexing, versioning, and failure recovery
Module 2: Advanced document preparation and chunking
- Comparing fixed size, recursive, semantic, and structure aware chunking approaches
- Choosing chunk size and overlap based on document structure and expected query patterns
- Preserving headings, document hierarchy, and source metadata during ingestion
- Handling tables, code, long sections, duplicate content, and poorly structured documents
Module 3: Embeddings and vector index design
- Selecting embedding models based on language, domain, dimensionality, cost, and quality
- Understanding similarity metrics and their effect on retrieval behavior
- Designing vector indexes and metadata fields for scalable filtering and retrieval
- Managing embedding updates when source data or embedding models change
Module 4: Metadata filtering and retrieval architecture
- Designing metadata schemas for permissions, document types, dates, categories, and ownership
- Applying filters before or during retrieval to narrow the candidate set
- Combining semantic similarity with business rules and structured constraints
- Preventing irrelevant or unauthorized content from entering the generation context
Module 5: Hybrid retrieval and search strategies
- Combining vector similarity with keyword and full text search
- Understanding why exact identifiers and domain terminology can require lexical retrieval
- Merging results from multiple retrieval strategies into a common candidate set
- Tuning candidate counts and retrieval depth for quality, latency, and cost
Module 6: Reranking and relevance improvement
- Applying a second ranking stage to improve the ordering of retrieved candidates
- Comparing model based reranking, semantic ranking, and simpler scoring approaches
- Selecting an appropriate number of candidates for retrieval and final context
- Balancing improved relevance against additional latency and processing cost
Module 7: Query transformation and context construction
- Rewriting ambiguous or conversational questions into clearer retrieval queries
- Using decomposition for questions that require information from several sources
- Building context from the highest value passages while respecting token limits
- Removing duplicate, contradictory, or low relevance content before generation
Module 8: RAG evaluation and systematic improvement
- Separating retrieval quality from generation quality during evaluation
- Measuring whether relevant evidence was retrieved before evaluating the final answer
- Creating representative test questions with expected sources or answer criteria
- Using evaluation results to improve chunking, retrieval, reranking, prompts, and model settings
Module 9: Grounded generation, citations, and answer control
- Designing prompts that instruct the model to prioritize retrieved evidence
- Generating source references and citations from retrieved metadata
- Handling conflicting sources and insufficient evidence explicitly
- Designing abstention behavior when the available context does not support a reliable answer
Module 10: Security and access controlled retrieval
- Applying document level and user level permissions during retrieval
- Propagating identity and authorization information through the RAG pipeline
- Protecting sensitive information in indexes, prompts, logs, and generated responses
- Considering prompt injection and malicious instructions embedded in retrieved documents
Module 11: Observability, performance, and cost optimization
- Logging queries, retrieved documents, scores, model responses, and execution metadata
- Tracing failures across ingestion, retrieval, reranking, and generation stages
- Measuring latency and identifying expensive stages in the RAG pipeline
- Balancing model quality, context size, retrieval depth, caching, and infrastructure cost
Module 12: Deployment, maintenance, and advanced retrieval patterns
- Designing separate development, test, and production configurations
- Automating document ingestion, incremental updates, reindexing, and validation
- Introducing multi source routing and agentic retrieval for more complex information needs
- Building a practical production readiness checklist covering quality, security, monitoring, ownership, and maintenance