Researcher at Microsoft Research Montréal working on coding agents, LLM agent training, and verifiable SWE benchmarks
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
24
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—
No output score is computed: the snapshot contains raw counts (recent3yWorks 23; yearly counts through 2026), but Evolve has not derived a normalized productivity score against comparable researchers, so this axis is left unscored.
Impact—
No impact score is computed: although the snapshot reports an h-index of 24 and an i10-index of 33, citation-based impact for this axis has not been normalized or verified, so no estimate is offered here.
Focus0/100
Focus scores 0, the radar's floor, reflecting breadth rather than a narrow specialty: the OpenAlex topic profile spans five clusters — Topic Modeling (40), Natural Language Processing Techniques (31), AI-based Problem Solving and Planning (25), Reinforcement Learning in Robotics (20), and Artificial Intelligence in Games (15) — while the current field distribution splits evenly, 0.5 versus 0.5, between coding agents and ML/RL for software engineering.
Collaboration60/100
Evolve's collaboration score is 60. The snapshot records 19 distinct coauthors, and 12 frequent collaborators — including Alessandro Sordoni, Marc-Alexandre Côté, Zhengyan Shi, Young Jin Kim, and Emiliano Penaloza — point to a stable, Montréal-centred working group rather than one-off partnerships. Both reviewed 2026 papers are middle-author, non-corresponding team efforts, consistent with a group-scale experimental culture at Microsoft Research Montréal.
Activity93/100
Activity scores 93, the strongest axis on this radar. The counts-by-year record shows 9 indexed works in 2023 and 9 in 2024, 6 in 2025, and 8 already in 2026, with recent3yWorks totalling 23 — an unusually dense three-year stretch. Output has been recorded every year from 2011 through 2026, marking the researcher as highly active in the current arXiv-driven coding-agent cycle.
Reproducibility38/100
Reproducibility scores 38 (reproMean), computed from 2 assessed papers in the catalog. ProgramDistill is rated low reproducibility readiness — strong engineering but no public release yet — while FrogNano is rated medium, with Evolve flagging validation-set reuse, base-score discrepancies, and missing significance tests. Neither reviewed preprint currently clears a high release-and-evidence bar, so artifact availability is the main limiting factor.
Research summary
Xingdi Yuan is a researcher at Microsoft Research Montréal whose profile is name-resolved from OpenAlex bibliographic data and centers on AI for software engineering and machine learning for LLM agents: the current field distribution splits evenly (0.5 and 0.5) between coding agents and reinforcement-learning/LLM approaches for software engineering. The record shows sustained, rising productivity — the radar's activity score is 93, with 23 works in the trailing three years, 9 indexed works in each of 2020, 2023, and 2024, and 8 already in 2026. Earlier topic counts span five clusters — Topic Modeling (40), Natural Language Processing Techniques (31), AI-based Problem Solving and Planning (25), Reinforcement Learning in Robotics (20), and Artificial Intelligence in Games (15) — consistent with a focus score of 0, i.e., broad coverage rather than a single narrow lane. Collaboration scores 60, with 19 distinct coauthors and a recurring Montréal-centred group. Reproducibility scores 38 across 2 assessed papers: ProgramDistill is rated low reproducibility readiness, and FrogNano medium, with Evolve citing thin statistics, validation-set reuse, and no public release. Output and impact dimension scores are not yet computed, so none are reported.
Collaboration pattern
Collaboration is team-based and Montréal-centred: the snapshot lists 19 distinct coauthors, with a recurring core including Alessandro Sordoni, Marc-Alexandre Côté, Zhengyan Shi, Young Jin Kim, Emiliano Penaloza, Jeonghye Kim, and Minseon Kim. Both reviewed 2026 papers are middle-author, non-corresponding contributions, suggesting large internal coding-agent project teams at Microsoft Research Montréal rather than solo-led efforts.
Research keywords
Highlights
ProgramDistill: 26 web apps become 4,063 replay-verified SWE agent tasks
FrogNano: online task synthesis lifts a 4B model to 61.5% on SWE-bench Verified
Timeline
- 2011Earliest year in the OpenAlex counts-by-year record, with 1 indexed work.
- 2017Indexed output rose to 6 works, up from 4 in 2016.
- 2020A peak year with 9 indexed works.
- 20239 indexed works, tying 2020 and 2024 as the record's highest years.
- 20268 works recorded so far, including the two reviewed arXiv preprints, ProgramDistill and FrogNano.
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.
- Bibliometrics come from an OpenAlex snapshot only; the h-index (24) and i10-index (33) are as reported and not independently verified.
- Output and impact dimension scores are not computed in the radar (marked as missing data), so no productivity or citation-impact score appears in this profile.
- The input metadata lists country 'US' while the affiliation is Microsoft Research Montréal, Canada; the affiliation is shown as provided and the mismatch was not reconciled.
- Both reviewed papers are 2026 arXiv preprints with limited independent validation; the profile is name-resolved from public bibliographic data.