Microsoft researcher bridging software engineering, cloud systems, and LLM distillation for agents and post-training
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
20
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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 available yet. This axis would measure publication volume and productivity over time; partial raw signals exist (i10-index 34; 35 works recorded in 2024) but Evolve has not converted them into an axis score.
Impact—
No impact score is available yet. This axis would measure citation-based influence such as citations per paper and field-weighted reach; the h-index of 20 exists as a raw signal but has not been converted into an axis score.
Focus0/100
Focus scores 0, the lowest value on Evolve's 0-100 scale. Topic tagging spreads across five OpenAlex areas: Software System Performance and Reliability (61 works), Cloud Computing and Resource Management (54), Software Engineering Research (53), Topic Modeling (30), and Software Engineering Techniques and Practices (25). The reviewed papers also split evenly, 0.5 and 0.5, between LLM distillation and LLM agents and memory, so no single field dominates the record.
Collaboration54/100
Collaboration scores 54. The OpenAlex snapshot lists 14 distinct coauthors across the most recent three years, a moderate-size, repeated-partner network. The frequent-coauthor pool centers on the LLM work (Xuchao Zhang, Menglin Xia, Yuxiao Yang, Tianrun Yu, Shangzhe Li, Kaixiang Zhao, Huaxiu Yao, Taylor W. Killian, Weitong Zhang, Pirzada Suhail), and both reviewed 2026 papers place the researcher in a middle author position without corresponding role, consistent with team-based contribution.
Activity100/100
Activity scores 100, the top band on the radar. The counts-by-year series shows 89 works across the most recent three years, with 35 works in 2024, 24 in 2025, and 30 already in 2026, a steep rise from 7 works in 2019 and 10 in 2020. Recent output is both sustained and accelerating relative to the earlier career window in the snapshot.
Reproducibility55/100
Reproducibility scores 55, computed from 2 reviewed papers with a mean repro-readiness of 55. Both carry a medium rating: PSD describes distillation pipelines around GPT-4.1-mini and Qwen3, but its LoCoMo results lack significance tests, while the EOS-token study spans three model families yet remains small-scale. Artifacts look partially reusable rather than turnkey.
Research summary
Chetan Bansal is a researcher at Microsoft (United States) whose bibliometric record spans software engineering, cloud systems, and, most recently, machine learning for large language models. The OpenAlex-resolved snapshot shows an h-index of 20 and an i10-index of 34, with activity concentrated in recent years: 89 works in the last three years, including 35 works in 2024, 24 in 2025, and 30 already in 2026. Topic tagging places 61 works under Software System Performance and Reliability, 54 under Cloud Computing and Resource Management, and 53 under Software Engineering Research, with Topic Modeling (30) and Software Engineering Techniques and Practices (25) rounding out a deliberately broad record — a spread that yields a focus score of 0 on Evolve's radar. The two papers Evolve has reviewed, both 2026 arXiv preprints with middle-author positions, sit in LLM post-training: PSD introduces pseudo self-distillation to transfer memory-representation capabilities into LLM agents, while the EOS-token study traces length inflation in on-policy distillation to student-teacher termination mismatch and proposes probability-level alignment across three model families. Both carry medium repro-readiness, and Evolve's verdicts flag missing significance tests in the former and small scale in the latter.
Collaboration pattern
Evolve's radar scores collaboration at 54. The OpenAlex snapshot lists 14 distinct coauthors across the most recent three years, a moderate-size, repeated-partner network. The frequent-coauthor pool centers on the LLM work, with Xuchao Zhang, Menglin Xia, Yuxiao Yang, Huaxiu Yao, Taylor W. Killian, and Weitong Zhang among the recurring names; both reviewed 2026 papers use middle-author, non-corresponding placements, consistent with team-based contribution.
Research keywords
Highlights
Pseudo self-distillation transfers agent-memory skill into smaller models
Evaluation caveats temper the PSD memory benchmark story
EOS-token mismatch explains length inflation in on-policy distillation
Timeline
- 2009Earliest publication year recorded in the OpenAlex snapshot, with 1 work
- 2019Output picks up: 7 recorded works after earlier single-digit years
- 202010 recorded works as cloud and software engineering themes take hold
- 2024Peak recorded output: 35 works in a single year
- 2026Two Evolve-reviewed arXiv preprints on LLM distillation and agent memory, both in middle-author position
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 metrics are computed from OpenAlex records only; output and impact axis scores are not yet available, so citation-based influence is not fully represented in this profile.
- Only two papers have completed Evolve review, both 2026 arXiv preprints; peer-reviewed venue status is not reflected in this profile.
- The frequent-coauthor list contains duplicate entries and includes the profile subject himself, so partner rankings are indicative rather than exact.
- Chetan Bansal is a fairly common name; identity is name-resolved from public bibliographic data (OpenAlex, ORCID 0000-0003-0102-8139) and records may be merged or split.