Back to Research Notes Index
UW / Allen Institute // 2024TOPIC: RAGTechnical Note
Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection
AUTHORS: Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, Hannaneh Hajishirzi | PUBLISHED & REVIEWED: 2025-02-08
CORE ARCHITECTURAL THESIS & FINDING
Introduces reflection tokens that allow models to dynamically decide when to retrieve passages, critique retrieved relevance, and evaluate factual groundedness.
MY INTERPRETATION
Gating retrieval behind confidence thresholds avoids introducing irrelevant noise into context windows on simple conversational turns.
PRACTICAL IMPLEMENTATION
Developed adaptive retrieval triggers in SourceLatch and PortfolioChatbot that distinguish between general conversational queries and knowledge-seeking queries.
RETRIEVAL_INTELLIGENCE // HYBRID_RAGVECTOR RETRIEVAL & RERANKING
STEP 01
Chunking
512 token splits
STEP 02
Embedding
Dense vectors
STEP 03
Qdrant Search
Cosine sim (k=25)
STEP 04
Cross-Encoder
Rerank top-5
STEP 05
Grounded Gen
With citations
SYSTEM LIMITATIONS, RUNTIME OVERHEAD & PRODUCTION CONSTRAINTS
- •Requires fine-tuned reflection classifiers or structured system prompt guidance.
EVIDENCE & REPRODUCIBILITY METHODOLOGY
Analysis of arXiv:2310.11511 empirical evaluation datasets.