Syllabus: RAG and Optimization
1. Header Information
| Field | Value |
|---|---|
| Technical Group | ? |
| Topic Name | RAG and Optimization |
| Topic Code | ? |
| Version | 1.0 |
| Training Audience | Freshers |
2. Course Objectives
[!NOTE] This topic introduces advanced techniques and optimizations for Retrieval-Augmented Generation (RAG) systems; adapting trainees with skills, lessons, and practices specific to AI applications, enterprise vector search, and GraphRAG knowledge bases.
Output Standards table
| Name | Code | Description |
|---|---|---|
| Advanced Indexing | ? | Understand Semantic Chunking vs Recursive Chunking and optimize Vector DBs using HNSW. |
| Hybrid Search | ? | Combine traditional BM25 keyword search with Vector Search using Reciprocal Rank Fusion (RRF). |
| Query Transformation | ? | Utilize LLM capabilities for HyDE and Query Decomposition to augment queries before retrieval. |
| Post-Retrieval Processing | ? | Filter and re-rank search results using Cross-Encoder models and Maximal Marginal Relevance (MMR). |
| GraphRAG Implementation | ? | Build sophisticated hybrid RAG architectures leveraging Knowledge Graphs and Neo4j. |
Detailed Learning Outcomes (LOs)
?LO1- Able to comprehend and apply Semantic Chunking vs Recursive Chunking.?LO2- Able to optimize vector databases using the HNSW Index (M, ef_construction, ef_search).?LO3- Able to leverage the BM25 algorithm alongside Vector Search.?LO4- Able to merge multi-search results using Reciprocal Rank Fusion (RRF).?LO5- Able to implement Hypothetical Document Embeddings (HyDE).?LO6- Able to perform Query Decomposition for complex multi-intent queries.?LO7- Able to utilize Cross-Encoder for enhanced Re-ranking vs Bi-Encoder approaches.?LO8- Able to apply Maximal Marginal Relevance (MMR) for diverse retrieval.?LO9- Able to design Neo4j graph schemas, extract entities, and maintain a knowledge graph.?LO10- Able to execute Graph Cypher QA Chains with Neo4j.
3. Topic Outline
- Unit 01: Advanced Indexing
- Unit 02: Hybrid Search
- Unit 03: Query Transformation
- Unit 04: Post-Retrieval Processing
- Unit 05: GraphRAG Implementation
4. Time Allocation
| Activity Type | Percentage | Description |
|---|---|---|
| Concept/Lecture | 40% | Concepts, theory (4 Sessions) |
| Assignment/Lab | 40% | Assignment, Lab |
| Guides/Review | 10% | Self review/Cross review |
| Exam | 10% | Pre-Test, Final Topic Test |
5. Training Materials & Environments
Textbooks
- Advanced AI Module: RAG and Optimization Guidelines
- Sample Data: FSoft_HR.pdf
References
- LangChain Documentation
- Neo4j Documentation
- ChromaDB Documentation
- Milvus Documentation
- Qdrant Documentation
- OpenAI Documentation
- Docling Documentation
Technical Requirements
- Python 3.10+
- LangChain, LangGraph
- Neo4j Database
- IDE (VSCode / PyCharm)
6. Assessment Scheme
| Component | Quantity | Weight (%) | Note |
|---|---|---|---|
| Quiz | 5 | 20% | In-class labs |
| Assignments | 2 | 30% | In-class labs |
| Final Practice Test | 1 | 50% | Comprehensive project |
Pass Criteria
- Total topic GPA >= 7.0/10
- Completed 100% Online video, Quiz, Assignments, Labs, Final Test
- Succesful demo of the Final Project (RAG system with advance function)
7. Training Delivery Principles
| Role | Responsibility/Criteria |
|---|---|
| Trainees | Must be familiar with basic RAG. |
| Trainer | Certified AI Engineer or higher. |
| Training | Daily offline training and labs. |
| Re-Test | Allowed 1 time for Failed status. |
| Marking | Direct grading of Practice tests. |
| Waiver Criteria | Equivalent industry certificates. |
| Others | Need personal laptop for labs. |