Oxford researcher on self-improving medical AI agents and the safety of LLMs in clinical decision-making
2
papers here
1
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
1
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—
Output volume cannot be scored yet: OpenAlex reports only an h-index of 1 and i10-index of 0, too sparse to rank productivity. This axis would measure total publications and preprints, weighted by venue and role.
Impact—
Impact cannot be scored yet: an i10-index of 0 indicates no papers with 10+ citations in the indexed window, and citation counts are not yet supplied. This axis would measure citations, uptake, and downstream influence of the work.
Focus100/100
Focus scores 100, the strongest axis in this profile. Evolve's field attribution places the reviewed output entirely within medical artificial intelligence (agentic systems), and OpenAlex topics are led by 'Artificial Intelligence in Healthcare and Education' (4 records) and 'Explainable Artificial Intelligence' (2). The two 2026 catalog entries — a medical RSI framework and an LLM-persuasion study — stay on one clinical-AI line, though quantum-computing topic tags suggest some breadth or namesake records.
Collaboration39/100
Collaboration scores 39, a modest level. The OpenAlex record lists 6 distinct coauthors, with repeated work within a small recurring circle: Jiayuan Zhu, Fenglin Liu, Minghao Hu, Jiazhen Pan, and Minhao Hu (the latter two names may be spelling variants of one collaborator). On the reviewed papers Junde Wu shifts position across the author list — first author on MedRSI and last author on the clinical-persuasion study — indicating both hands-on and senior roles within a compact team.
Activity52/100
Activity scores 52. OpenAlex records 5 works in the most recent three-year window, all dated 2026 in counts-by-year — an intense but very recent burst rather than a sustained multi-year record. The Evolve catalog mirrors this: both reviewed papers appeared in 2026, including MedRSI as first author. With only one publication year represented, the score reflects an emerging researcher whose productivity is concentrated in a single active year and not yet testable over time.
Reproducibility55/100
Reproducibility scores 55, based on a single assessed paper (reproMean 55 from 1 paper), so the signal is preliminary. MedRSI carries a 'medium' repro-readiness: Evolve notes strong benchmark gains but flags LLM-judged clinical cost and private data as limits on independent verification. No other cataloged output has a repro assessment, leaving the axis thin; releases with public data and code would be the natural way to strengthen this score.
Research summary
Junde Wu is a researcher at the University of Oxford working at the intersection of medical artificial intelligence and agentic LLM systems. The profile is name-resolved from OpenAlex bibliographic data and is at an early stage: the indexed record shows an h-index of 1, an i10-index of 0, and 5 works within the most recent three-year window, all dated 2026. The Evolve catalog anchors the picture on two 2026 arXiv papers. The first, MedRSI, has Junde Wu as first author and claims the first recursive self-improvement framework for medical agents, using clinically aligned self-evolution so agents can create and refine their own tools, with reported benchmark gains on tasks such as glaucoma diagnosis and echocardiography; Evolve rates its repro-readiness as medium, noting that LLM-judged clinical cost and private data limit verification. The second, 'AI Can Be Easily Persuaded in Clinical Decision Making,' places Junde Wu in the last author position and probes a safety weakness of clinical AI. Topic tags center on 'Artificial Intelligence in Healthcare and Education' (4) and 'Explainable Artificial Intelligence' (2), with single records in quantum mechanics, quantum computing, and AI ethics — the quantum entries may reflect namesake records. Collaboration is compact, with 6 distinct coauthors including Jiayuan Zhu, Fenglin Liu, Minghao Hu, Jiazhen Pan, and Minhao Hu. Focus is the standout dimension at 100; impact and output cannot yet be scored.
Collaboration pattern
Collaboration is compact: the OpenAlex record shows 6 distinct coauthors, and the recurring names are Jiayuan Zhu, Fenglin Liu, Minghao Hu, Jiazhen Pan, and Minhao Hu (possibly two spellings of one person). Author positions flip between first (MedRSI, 2026) and last (the 2026 clinical-persuasion study), suggesting this researcher both leads projects and anchors senior supervision within a small team. No large multi-institution network is visible in the current data.
Research keywords
Highlights
Claims the first recursive self-improvement framework for medical agents
Early evidence on how easily clinical AI is persuaded
In 'AI Can Be Easily Persuaded in Clinical Decision Making' (arXiv 2026), Junde Wu is the last author, examining how AI systems involved in clinical decisions can be swayed — a safety question that complements the clinical-alignment agenda of MedRSI. This paper is not yet reviewed by Evolve and has no keywords, slug, or repro assessment in the catalog.
A tightly focused but early-stage clinical-AI agenda
Timeline
- 2026MedRSI appears on arXiv: Junde Wu is first author of a claimed-first recursive self-improvement framework for medical agents, with medium repro-readiness in Evolve's assessment.
- 2026'AI Can Be Easily Persuaded in Clinical Decision Making' is posted to arXiv with Junde Wu in the last author position, extending the clinical-safety line; it is not yet reviewed by Evolve.
- 2026OpenAlex records 5 works in the recent three-year window, all in 2026, alongside an h-index of 1 and i10-index of 0 — an early-stage, single-year publication burst.
Papers in the Evolve catalog
Frequent co-authors
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.
- Bibliometric scores come from OpenAlex only (identity source: openalex); h-index 1 and i10-index 0 reflect a very early record and may lag actual citations.
- The OpenAlex profile mixes topic tags outside medical AI — quantum mechanics and quantum computing algorithms and architecture — so namesake records may be aggregated under this name; 'Junde Wu' is name-resolved from public bibliographic data and no ORCID link was available.
- Only 2 papers from the Evolve catalog back this profile; the second has no Evolve verdict, keywords, slug, or repro assessment yet.
- Reproducibility and impact rest on one assessed paper and sparse citation data; treat all dimensions as early-stage results.