Part III: Engineering Intelligent Systems
RAG pipeline, best practices, and agent architectures.
The RAG Pipeline
Explore the main stages of RAG.
🗂️
1. Indexing
Chunking, embeddings, and vector DB indexing.
🔍
2. Retrieval
Similarity search finds relevant chunks.
✍️
3. Generation
Augment the prompt and generate grounded answers.
RAG Best Practices
Chunking
Semantic or heading-aware chunking with modest overlap.
Hybrid Search
Dense vectors + BM25 for exact terms.
Reranking
Cross-encoder rerank top-K results.
Filters
Use metadata to narrow scope.
Citations
Return source URLs/IDs with spans.
Evaluation
Measure precision/recall and faithfulness.
AI Agents
Components
- Planner
- Tools & Environment
- Memory & State
Planning & Tool Use
- Decompose tasks
- Validate IO
- Fallbacks
Memory Systems
- Short-term context
- Long-term vector store
- Episodic artifacts