Carnegie Mellon researcher bridging LLM agent systems efficiency and open, verifiable clinical AI benchmarking
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
17
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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—
No scored output axis is available in this snapshot beyond per-year OpenAlex work counts; verified venue tiers and authorship shares are lacking. This axis would measure total research volume, outlet quality, and first-/last-author share across the full record.
Impact—
Impact is not scored here: the snapshot supplies only aggregate h-index (17) and i10-index (20) without citation distributions, per-paper citations, or venue-level influence data. A scored impact axis would measure citation reach and downstream adoption of this researcher's work.
Focus0/100
The focus score is 0, reflecting a record spread across two equally weighted fields — machine learning (LLM agents and systems) at 0.5 and clinical AI / medical informatics (LLM benchmarking) at 0.5 — plus five OpenAlex topic clusters led by Topic Modeling (23) and Natural Language Processing Techniques (15), with an anomalous Ferroelectric and Negative Capacitance Devices cluster (7). Read charitably this is two parallel lines rather than scattered work; the metric reads it as low concentration.
Collaboration39/100
The collaboration score is 39, modest against a small active network: 6 distinct coauthors in the last three years, with the frequent-coauthor list naming Trang Nguyen, Eulrang Cho, Bingqing Chen, Christine Park, and Valerie Chen. On both cataloged 2026 papers he sits in the last-author slot without corresponding-author designation, consistent with a senior supervisory role within compact teams over the 10 recent works.
Activity70/100
The activity score is 70, backed by a clear ramp in OpenAlex per-year counts: 1 work in 2015, intermittent output through 2021 (1-2 per year), then 7 works in 2022 and a peak of 13 in 2023 — the highest single-year figure in the snapshot — before 5 in 2024 and 5 already recorded for 2026. The recent3yWorks figure of 10 and two 2026 arXiv submissions corroborate sustained current activity.
Reproducibility78/100
The reproducibility score is 78, computed as reproMean 78 across the 2 papers Evolve has reviewed. The synthetic-EHR benchmark paper is rated high for reproducibility readiness — open benchmark, traceable labels, physician validation — while CliffCompaction is rated medium, with strong KernelBench results but thin statistical testing and baseline fairness. With only two reviewed papers, this axis is directionally useful but rests on a small base.
Research summary
Tim Dettmers is a researcher at Carnegie Mellon University whose recent output, as indexed by OpenAlex, splits evenly between two lines of work: machine learning for LLM agents and systems, and clinical AI / medical informatics focused on LLM benchmarking. The snapshot records an h-index of 17 and an i10-index of 20, with 10 works in the last three years and a clear activity ramp — from 7 works in 2022 to a peak of 13 in 2023, then 5 in 2024 and 5 already recorded for 2026. Within the Evolve review catalog, both of his 2026 arXiv papers place him in the last-author position without corresponding-author designation: 'Synthetic Hospital', an open, physician-validated longitudinal EHR benchmark built on FHIR R4 and knowledge-graph structures (rated high for reproducibility readiness), and 'CliffCompaction', a truncation-only context-compaction method reported to cut coding-agent cost by roughly 50% at maintained performance (rated medium). Across the two reviewed papers the reproducibility mean is 78. The record carries some resolution risk: the OpenAlex topic profile includes an unrelated 'Ferroelectric and Negative Capacitance Devices' cluster (7 works), and the researcher's own name appears in his frequent-coauthor list.
Collaboration pattern
Collaboration is concentrated in a small circle: 6 distinct coauthors in the last three years, with Trang Nguyen, Eulrang Cho, Bingqing Chen, Christine Park, and Valerie Chen the recurring names in the frequent-coauthor list. On both cataloged 2026 papers he occupies the last-author slot without corresponding-author designation, suggesting a senior supervisory position within compact teams. The collaboration dimension scores 39, consistent with few but stable partners rather than a broad, rotating coauthor network.
Research keywords
Highlights
Open, physician-validated synthetic EHR benchmark
Truncation-only compaction halves coding-agent cost
Timeline
- 2015First work recorded in the OpenAlex snapshot (1 work); sparse output of 1-2 works per year follows through 2021.
- 2022Output ramps to 7 works, the first sustained jump visible in the record.
- 2023Peak year: 13 works recorded, the highest single-year count in the snapshot.
- 20245 works recorded, consolidating after the 2023 peak.
- 20265 works recorded so far, including both Evolve-reviewed arXiv papers — 'Synthetic Hospital' and 'CliffCompaction' — with him as last author on each.
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.
- All bibliometrics come from a single OpenAlex snapshot; no ORCID or country field was available, so this profile is name-resolved from public bibliographic data and carries entity-resolution risk.
- The frequent-coauthor list includes the researcher's own name ('Tim Dettmers'), and the topic profile contains an unrelated 'Ferroelectric and Negative Capacitance Devices' cluster (7 works) — both signs of possible identity or topic-resolution noise.
- The output and impact axes are unscored; citation evidence in this snapshot is limited to h-index 17 and i10-index 20, with no per-paper citations, venue tiers, or citation distributions.
- Qualitative judgments rest on only 2 Evolve-reviewed papers (both 2026 arXiv preprints, last-author, non-corresponding); the reproducibility mean of 78 is computed over that small base.