Bowei He

Mohamed bin Zayed University of Artificial Intelligence · Hong Kong SAR China

Unclaimed · AI-structured profileOpenAlex

Early-career MBZUAI researcher on LLM agent memory, personalization, and self-improving agents

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-17) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

Not yet scored: cumulative output metrics (total publications, venue mix, first-author share) are insufficient in the current OpenAlex snapshot. This axis would measure sustained publication volume and authorship roles once full-indexed records accumulate.

Impact—

Not yet scored: citation-based impact cannot be meaningfully computed — the snapshot shows h-index 1 and i10-index 0 on very recently indexed work. This axis would measure citation uptake, influential papers, and field-level recognition over time.

Focus100/100

Scored at the maximum: the topic profile is tightly concentrated — Topic Modeling (13 records), Information Retrieval and Search Behavior (10), and Natural Language Processing Techniques (9) form the top three of five indexed topics, and the sole field assignment is natural language processing (LLM agent memory systems). All three catalogued 2026 papers — agent memory, web agents, autonomous research loops — sit inside one agent-centric NLP agenda.

Collaboration51/100

Mid-range score: the recent three-year window lists 12 distinct co-authors, with recurring partners including Xiaokun Zhang, Meng Ding, Xue Liu, Qixuan Sun and Yue Que. Authorship is mixed — first author on the SCAFFOLD and Beneath the Diff preprints, middle author on CreaMem — indicating an early pattern of team-based work that has not yet widened into a large independent network.

Activity100/100

Scored at the maximum: the OpenAlex snapshot records 30 works in the last three years, accelerating from 1 indexed work in 2025 to 29 in 2026. All three catalogued papers are 2026 arXiv preprints, two as first author. The trajectory indicates an intensive current production phase rather than an established multi-year record, so sustained pace remains to be confirmed.

Reproducibility100/100

Scored at the maximum, but on a narrow base: reproducibility is currently computed from one reviewed paper (reproPapers 1, reproMean 100). CreaMem carries a high reproducibility-readiness rating, consistent with an arXiv system paper built on public benchmarks (LoCoMo, LongMemEval-S). The two remaining 2026 preprints have not been assessed, so this score should be read as indicative rather than established.

Research summary

Bowei He is an early-career researcher at the Mohamed bin Zayed University of Artificial Intelligence whose public bibliographic footprint — resolved from OpenAlex records, as no ORCID is linked — centers on large language model agents, their memory systems, and the retrieval and topic-modeling techniques beneath them. The OpenAlex snapshot lists 30 works in the last three years, nearly all concentrated in 2026 (29 records versus 1 in 2025), with topic counts led by Topic Modeling (13), Information Retrieval and Search Behavior (10), and Natural Language Processing Techniques (9). Evolve's catalog holds three 2026 arXiv papers. CreaMem, a scene-aware memory architecture that partitions agent memory by life scenes and dual-codes entries, is reviewed in detail: it tops LoCoMo at 54.61% and LongMemEval-S at 66.40%, with high reproducibility readiness, though its headline gains lack significance testing. Two further first-authored preprints, SCAFFOLD on self-improving web agents via recursive parametric skill abstraction and Beneath the Diff on algorithmic mode collapse in code-level autonomous research loops, are indexed but not yet reviewed. Citation indicators remain embryonic (h-index 1, i10-index 0), and collaboration is team-based, with 12 distinct recent co-authors including Xiaokun Zhang and Meng Ding. This is an early-stage profile; the output and impact axes are not scored.

Collaboration pattern

The recent three-year window shows 12 distinct co-authors, with a stable core including Xiaokun Zhang, Meng Ding, Xue Liu, Qixuan Sun, Yue Que and Jin Guo, among others. The pattern is team-based rather than solo or broadly networked: He is first author on two 2026 preprints (SCAFFOLD; Beneath the Diff) and a middle author on the reviewed CreaMem paper, consistent with a collaboration score of 51 at this career stage.

Research keywords

LLM agentslong-term memorymemory architectureepisodic memoryretrieval-augmented generationpersonalizationtopic modelinginformation retrievalweb agentsskill abstractionself-improving agentsautonomous research loops

Highlights

CreaMem: scene-aware memory for personalized agents
Partitions agent memory by life scenes and dual-codes entries; Evolve's review reports leading results on LoCoMo (54.61%) and LongMemEval-S (66.40%) with high reproducibility readiness, while noting that the main gains lack significance testing.
SCAFFOLD: recursive skill abstraction for self-improving web agents
First-authored 2026 arXiv preprint proposing recursive parametric skill abstraction so that web agents can improve themselves; indexed in the Evolve catalog but not yet reviewed, so no field or reproducibility metadata is attached.
Beneath the Diff: mode collapse in autonomous research loops
First-authored 2026 arXiv preprint diagnosing and mitigating algorithmic mode collapse in code-level autonomous research loops; not yet reviewed by Evolve, so it carries no verdict, keywords, or reproducibility rating.

Timeline

2025First indexed work appears in the OpenAlex record, with only 1 work dated 2025.
2026First-authored preprint SCAFFOLD is released, proposing recursive parametric skill abstraction for self-improving web agents.
2026First-authored preprint Beneath the Diff is released, diagnosing and mitigating algorithmic mode collapse in code-level autonomous research loops.
2026Joins CreaMem, a scene-aware dual-coded memory architecture for personalized agents, as a middle author; Evolve reviews it with high reproducibility readiness.
2026Output accelerates sharply: indexed works rise from 1 in 2025 to 29 in 2026.

Papers in the Evolve catalog

2026-09-08
CreaMem: A Scene-Aware Memory Architecture for Personalized Agents

Frequent co-authors

Xiaokun Zhang×1 Meng Ding×1 Xue Liu×1 Qixuan Sun×1 Yue Que×1 Jin Guo×1 Dihang Yang×1 Wenchang Situ×1 Chen Ma×1 Weixu Zhang×1 Yili Jin×1 Xue Liu×1

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

All bibliometrics come from an OpenAlex snapshot; no ORCID is linked, so this profile is name-resolved from public bibliographic data and "Bowei He" may be a common name.
Output and impact dimensions are unscored: citation indicators (h-index 1, i10-index 0) reflect recently indexed, early-stage work rather than a mature citation record.
Only one catalog paper (CreaMem) carries an Evolve review; SCAFFOLD and Beneath the Diff lack verdicts, keywords, and reproducibility ratings, so the reproducibility score rests on a single paper.
All three catalogued papers are 2026 arXiv preprints; the input data contains no peer-reviewed venue, grant, or award information.