Course Objective & Scope
Spinning up a basic vector-search demo over a single PDF takes thirty minutes, but running an enterprise knowledge assistant across thousands of policies, contracts, tables, and technical manuals requires rigorous engineering. This 1-2 day hands-on engineering lab guides technical teams through building, diagnosing, and optimizing a production-grade RAG (Retrieval-Augmented Generation) pipeline.
Participants implement layout-aware document parsing, structure-aware chunking, metadata enrichment, dense + lexical hybrid retrieval, cross-encoder reranking, grounded generation with strict inline citations, and automated evaluation suites that separate retrieval failures from generation failures.
Who Should Attend?
- AI/ML engineers and software developers building enterprise knowledge assistants and search-backed applications
- Data engineers, data scientists, and solution architects responsible for retrieval quality and vector infrastructure
- Technical teams troubleshooting hallucinations, poor recall, or broken table handling in existing RAG prototypes
Key Takeaways
- Parse messy PDFs, tables, and hierarchical documents cleanly and select structure-aware chunking strategies
- Combine dense embeddings, lexical search (BM25), metadata filters, and query transformation in a hybrid retrieval pipeline
- Boost top-k precision using cross-encoder reranking and parent-child context expansion
- Enforce grounded answer synthesis with verifiable inline citations and strict abstention when evidence is missing
- Run automated RAG evaluations measuring retrieval recall/precision alongside generation faithfulness and answer relevance
Curriculum
Production RAG Architecture and Failure Modes
- Why naive RAG fails on enterprise corpora: diagnosing retrieval misses vs. generation hallucinations
- The end-to-end architecture: Ingestion -> Indexing -> Query Rewriting -> Hybrid Search -> Reranking -> Synthesis
- Designing for document-level access control lists (ACLs) and incremental index updates
Document Parsing, Chunking, and Metadata Enrichment
- Layout-aware parsing for multi-column PDFs, nested tables, and policy hierarchies
- Benchmarking fixed-window, markdown-header, recursive, and parent-child chunking strategies
- Enriching chunks with structural breadcrumbs, effective dates, document types, and contextual headers
Embeddings, Hybrid Retrieval, and Query Transformation
- Selecting multilingual and domain-appropriate embedding models and vector index parameters
- Combining dense vector similarity with BM25 lexical search using Reciprocal Rank Fusion (RRF)
- Query transformation techniques: multi-query expansion, sub-question decomposition, and self-query metadata filtering
Reranking, Context Assembly, and Cited Generation
- Integrating cross-encoder rerankers to maximize precision in the final context window
- Prompting for strict grounding: handling conflicting document versions and enforcing abstention on missing evidence
- Generating structured responses with verifiable inline citations mapped to source chunks and page numbers
Systematic RAG Evaluation and Pipeline Tuning
- Building a golden evaluation dataset combining synthetic question generation with SME curation
- Measuring retrieval metrics (Recall@k, MRR, Context Precision) separately from generation metrics (Faithfulness, Relevance)
- Running ablation experiments across chunk sizes, top-k, and rerankers to hit production quality gates
Other Courses in This Category
-
A 1-2 day interactive tabletop simulation for executives and risk teams testing incident response, hallucination crises, data-leak scenarios, and EU AI Act compliance controls under pressure.
- Live Course
- 5 modules
-
A 0.5-1 day applied workshop on call transcript analytics, Voice of the Customer synthesis, intelligent intent routing, real-time agent-assist copilot design, and automated QA scoring.
- Live Course
- 4 modules
-
A 1-2 day sector-focused workshop applying AI and automation to demand forecasting, dynamic route and dispatch optimization, proactive exception alerts, and warehouse document processing.
- Live Course
- 5 modules
-
A 0.5-1 day hands-on workshop on designing metric trees, separating North Star, leading, and lagging indicators, refactoring cluttered executive dashboards, and establishing decision rituals.
- Live Course
- 5 modules
-
A 1-2 day immersive role-play and case simulation lab where managers practice high-stakes feedback, tough stakeholder negotiations, conflict resolution, and coaching conversations with live feedback.
- Live Course
- 5 modules
-
A 1-2 day hands-on workshop that maps departmental pain points into high-value AI use case candidates, scores feasibility versus ROI, and produces an actionable 90-day AI implementation roadmap.
- Live Course
- 5 modules
Get in touch
