This is my experimental implementation of some modern agentic RAG features that I have been reading about.
Specifically, I am aiming to implement:
- Tiered Document Understanding
- Layout-Aware Parsing
- Academic & Scientific Parsing
- High-Fidelity Multi-Modal Parsing
- Multi-Layered Indexing Strategy
- Semantic Chunking
- Propositional Indexing
- Graph Indexing
- Hierarchical Abstraction (RAPTOR)
- Two-Stage Retrieval
- Domain-Specific Embedding
- Late Interaction Re-ranking (ColBERT)
- LLM Constellation (Mixture-of-Experts)
- Programmatic Prompt Optimization (DSPy)
- Stateful Graph-Based Orchestration (LangGraph)
- Dynamic Multi-Step Planning (Tree of Thoughts - ToT)
- Iterative Refinement & Self-Critique (CRITIC Loop)
- Corrective-RAG (CRAG)
- Experiential Memory
- Context Distillation
- Vector Quantization
- Automated RAG Evaluation
- End-to-End Tracing
My understanding of these technologies is entry-level, so implementation is highly experimental.