PER: Shunyu Yao
PUMA Context
Overview
Author of ReAct (2022) — the foundational paper for PUMA’s Stage 4 agent architecture — and Tree of Thoughts (2023), relevant for Stage 3 backlog prioritisation.
Key Contribution
ReAct’s Thought-Action-Observation loop is the base pattern for all PUMA agent stages from Stage 4 onwards. The paper’s empirical demonstration that grounding LLM reasoning in external observations reduces hallucination is directly relevant to PUMA’s triage agent design.
Related notes:
- LN-Yao-2022-ReAct
- LN-Yao-2023-TreeOfThoughts
- PN-ReAct-AgentPattern — ReAct concept note
- PN-CoT-FewShot-Prompting — CoT contrasted with ReAct
- EX-Stages-Overview — Stage 4 RAG+ReAct
- MOC-LLM-Benchmarks-PM-AI