Microsoft Research Montréal researcher building verifiable task pipelines and RL-trained coding agents
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
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—
This axis measures publication volume and multi-year productivity. The underlying data is not yet available: OpenAlex logs only 3 works, all dated 2026, so no output trend can be computed or scored yet.
Impact—
Impact measures citation accumulation such as h-index and per-year citations. The snapshot reports h-index 0 and i10-index 0 on works all dated 2026, leaving no time to accrue citations, so this axis cannot be scored yet.
Focus0/100
The focus score is 0 on this snapshot. OpenAlex topics spread across five areas — software engineering research (2) and techniques (2), cryptography and data security (1), blockchain applications (1), and teaching and learning programming (1) — although the recent field distribution is tighter, splitting 0.5/0.5 between coding agents and LLM/RL for software engineering.
Collaboration60/100
Collaboration scores 60. The record shows 19 distinct coauthors, with 12 recurring collaborators listed, including Marc-Alexandre Côté, Alessandro Sordoni, Xingdi Yuan, and Zhengyan Shi — a pattern consistent with team-based industrial research. Both reviewed papers sit in middle-author positions on group efforts rather than solo or first-authored work.
Activity40/100
Activity scores 40. The counts-by-year record holds a single entry — 3 works in 2026 — matching the recent-3-year count of 3. Output is recent and clustered rather than sustained across years, so the moderate score reflects a short observable window rather than any decline.
Reproducibility38/100
Reproducibility scores 38, the mean over 2 assessed papers. ProgramDistill rates low readiness — strong engineering but no public release yet — while FrogNano rates medium, with a credible recipe tempered by validation-set reuse and missing significance testing. Neither reviewed paper currently ships fully reusable artifacts.
Research summary
Matheus Pereira is a researcher at Microsoft Research Montréal working at the intersection of AI for software engineering and machine learning for LLM agents. This profile is name-resolved from OpenAlex bibliographic data. The recorded output is recent and concentrated: 3 works logged in 2026, a recent-3-year count of 3, and citation indicators (h-index 0, i10-index 0) that simply reflect how new the record is. OpenAlex topic tags span software engineering research (2) and techniques (2), plus cryptography and data security, blockchain applications, and teaching and learning programming (1 each); the recent field distribution splits evenly (0.5/0.5) between coding agents and LLM/reinforcement-learning methods for software engineering. Two 2026 arXiv preprints reviewed by Evolve anchor the catalog. ProgramDistill converts 26 interactive web apps into 4,063 replay-verified SWE agent tasks, though with thin statistics and no public release yet. FrogNano trains a 4B coding agent with distillation-free reinforcement learning over roughly 1,500 synthetic tasks, reaching 61.5% on SWE-bench Verified, with caveats around validation-set reuse, base-score discrepancies, and missing significance testing. Collaboration is broad for the record's size: 19 distinct coauthors, with recurring partners including Marc-Alexandre Côté, Alessandro Sordoni, and Xingdi Yuan. Reproducibility readiness is mixed, averaging 38 across the 2 assessed papers.
Collaboration pattern
The record lists 19 distinct coauthors and 12 recurring collaborators — notably Marc-Alexandre Côté, Alessandro Sordoni, Xingdi Yuan, Zhengyan Shi, Roger Creus Castanyer, and Emiliano Penaloza. Both catalogued papers are middle-authored team efforts rather than solo work, consistent with group-based industrial research at Microsoft Research Montréal. The collaboration score of 60 reflects a moderately broad but concentrated network.
Research keywords
Highlights
ProgramDistill: 26 web apps become 4,063 replay-verified SWE agent tasks
FrogNano: distillation-free RL trains a competitive 4B coding agent
Timeline
- 2026ProgramDistill preprint posted to arXiv, presenting an automated pipeline that converts 26 interactive web apps into 4,063 replay-verified SWE agent tasks.
- 2026FrogNano preprint posted to arXiv, training a 4B coding agent via online task synthesis and reporting 61.5% on SWE-bench Verified.
- 2026OpenAlex logs 3 works in 2026 — the entire recorded output to date — with recent-3-year works also standing at 3.
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; no ORCID was available to verify identity, and 'Matheus Pereira' is a common name — treat this profile as name-resolved from public bibliographic data.
- Output and impact dimensions lack scorable data (only 3 works, all dated 2026; h-index 0 and i10-index 0), so no productivity or citation assessment is offered.
- Only 2 recorded works have been reviewed by Evolve, and both are arXiv preprints rather than peer-reviewed publications.
- The country field was not provided; affiliation is given as Microsoft Research Montréal (Canada).