Microsoft Research Montréal researcher building LLM coding agents and reinforcement learning for software engineering.
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
14
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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—
The output axis would measure total publication volume, per-year productivity, and venue quality. Evolve does not yet have enough validated publication records to score this axis for this profile, so no output score is reported.
Impact—
The impact axis would measure citation-based influence and downstream uptake of this researcher's work. Although the record lists an h-index of 14 and i10-index of 19, Evolve's impact score is not yet available and is not estimated here.
Focus0/100
Focus scores 0/100: the OpenAlex topic footprint is unusually broad, led by machine learning in materials science (13 works), battery materials (8), adversarial robustness (8), computational drug discovery (7), and data classification (6), while the reviewed work splits evenly (0.5/0.5) between coding-agent AI and LLM/RL for software engineering. This dispersion may partly reflect name-disambiguation noise.
Collaboration60/100
Collaboration scores 60/100: the record lists 19 distinct coauthors, with a dense recurring cluster of 12 frequent collaborators — including Alessandro Sordoni, Marc-Alexandre Côté, Xingdi Yuan, Zhengyan Shi, and Jeonghye Kim — consistent with team-based research at Microsoft Research Montréal rather than solo or two-person work. Both 2026 reviewed papers are multi-authored.
Activity87/100
Activity scores 87/100, the strongest axis: the publication record shows 15 works in 2023, 10 in 2024, 7 in 2025, and 2 already in 2026 — 19 works in the last three years. Output accelerated sharply after 2021 (5 works) and 2022 (7 works), versus sporadic records before 2018.
Reproducibility38/100
Reproducibility scores 38/100, computed across the 2 Evolve-reviewed papers. FrogNano is rated medium — the RL recipe is credible but validation-set reuse and missing significance testing temper the claims — while ProgramDistill is rated low, with strong engineering but no public release yet. Neither 2026 preprint currently meets a high release-and-evaluation bar.
Research summary
Minseon Kim is a researcher at Microsoft Research Montréal whose recent work centers on large language model agents for software engineering, especially coding agents and reinforcement learning for code. Evolve has reviewed two 2026 arXiv preprints from this record: FrogNano, a first-authored study reporting distillation-free RL over ~1,500 synthetic tasks that brings Qwen3.5-4B to 61.5% on SWE-bench Verified, and ProgramDistill, a middle-author contribution describing an automated pipeline that converts 26 web apps into 4,063 replay-verified agent tasks. The OpenAlex snapshot shows an active recent record: h-index 14, i10-index 19, 19 works in the last three years, and an accelerating curve — 15 works in 2023, 10 in 2024, 7 in 2025. Collaboration is team-based (19 distinct coauthors, 12 recurring, largely at Microsoft Research Montréal; score 60/100). Activity is the strongest axis at 87/100, while reproducibility scores 38/100 across the two reviewed papers (one medium, one low readiness). Focus scores 0/100 because the topic footprint is unusually broad — materials science, battery materials, adversarial robustness, drug discovery — which may partly reflect same-name disambiguation. Output and impact dimensions are not yet scored.
Collaboration pattern
Co-authorship is team-based: the record lists 19 distinct coauthors, and 12 recur frequently — a dense cluster around Microsoft Research Montréal including Alessandro Sordoni, Marc-Alexandre Côté, Xingdi Yuan, Zhengyan Shi, Jeonghye Kim, and Young Jin Kim. Both reviewed 2026 papers are multi-author efforts, with FrogNano first-authored. The collaboration score of 60/100 reflects a steady mid-sized network rather than a very large or scattered one.
Research keywords
Highlights
First-authored RL training turns a 4B model into a competitive coding agent
Automated pipeline converts 26 web apps into 4,063 verifiable SWE tasks
Reproducibility gap across both reviewed releases
Across the two reviewed 2026 papers, Evolve's reproducibility mean is 38/100 (2 papers): one medium, one low. Neither yet ships the public artifacts or statistical rigor Evolve expects, which is the main barrier to independent verification of this line of work.
Timeline
- 2003–2013Earliest records in the OpenAlex publication history are sporadic (1 work in 2003, 1 in 2005, 2 in 2010, 1 in 2013), predating the current research program.
- 2018–2020Publication resumes at low volume (1 work in 2018, then 2 per year in 2019 and 2020).
- 2021–2022Output accelerates: 5 works in 2021 and 7 in 2022, marking a sustained step-up in activity.
- 2023Peak year with 15 works in the publication record.
- 2024–2025High output is sustained with 10 works in 2024 and 7 in 2025.
- 2026Two arXiv preprints in agentic software engineering reviewed by Evolve: first-authored FrogNano (RL-trained 4B coding agent) and ProgramDistill (verifiable task generation from web apps).
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 an OpenAlex snapshot; the name 'Minseon Kim' is common, so the identity is name-resolved from public bibliographic data and the broad topic footprint may include same-name records.
- Evolve's output and impact dimensions are unscored for this profile; only h-index (14) and i10-index (19) are available, with no per-paper citation data.
- Only two papers have been reviewed by Evolve, both 2026 arXiv preprints; no peer-reviewed venue assessment is in scope.
- Institutional affiliation and career facts come from the input snapshot and are not independently verified.