Back to Research Notes Index
MIT // 2023TOPIC: AgentsTechnical Note
Reflexion: Language Agents with Verbal Reinforcement Learning
AUTHORS: Noah Shinn, Federico Cassano, Edward Berman, Ashwin Gopinath, Karthik Narasimhan, Shunyu Yao | PUBLISHED & REVIEWED: 2025-02-05
CORE ARCHITECTURAL THESIS & FINDING
Equipping autonomous language agents with verbal self-reflection memory of past trajectory failures enables rapid multi-step reasoning improvements without weight updates.
MY INTERPRETATION
Maintaining an episodic buffer of execution mistakes and tool errors allows agents to formulate self-correcting strategies on subsequent retries.
PRACTICAL IMPLEMENTATION
Integrated reflective critique nodes in LangGraph agent graphs to analyze failed API responses and rewrite payloads before escalating.
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
- •Self-reflection loops can risk circular reasoning without external ground-truth feedback.
EVIDENCE & REPRODUCIBILITY METHODOLOGY
Comparative benchmark review on HumanEval and AlfWorld agent benchmarks.