Publications

Publications and Preprints

My research focuses on interpretable AI systems: understanding and shaping the mechanisms behind LLM reasoning, agent memory, and human-centered recommendation. I am particularly interested in methods with clear mathematical or geometric structure that explain empirical phenomena, support verification, and make AI behavior more reliable and safe.

• SHELF: From Similarity Retrieval to Path-Aware Auditable Memory for LLM Agents
  Weihan Fei, advised by Prof. Xiang Wang
  Under Review at NeurIPS 2026
  – Path-aware auditable memory for LLM agents, with explicit retrieval paths, variable-level diagnosis, and targeted path revision.

Work in Progress

• Adaptive-Thinking for Generative Recommendation
  Weihan Fei, advised by Prof. An Zhang
  Ongoing research
  – Investigating when explicit reasoning helps or hurts generative recommendation, and whether think/not-think inference relies on different recommendation signals.

Selected Projects

• CoSwipe: AI Companion for Short-Video Feed Interaction
  Third Prize, Track 1, Douyin AI Innovator Plan 2026 Hackathon, USTC Station
  – An AI companion experience embedded in the short-video feed, designed to notice browsing behavior, respond like a friend, and make downstream content feel more personally connected. [Demo]