UT Austin researcher applying LLMs and programming-language semantics to code generation, editing, and evaluation
3
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
Output metrics are not yet scorable: the OpenAlex record links too few works (h-index and i10-index both 0) to measure publication volume, venue placement, or authorship role. This axis would track total scholarly output and productivity once more publications are linked to the profile.
Impact—
Impact cannot be scored yet: with zero indexed citations on the current OpenAlex cluster, there is no basis for citation-count, h-index, or influence estimation. This axis would measure citation uptake and scholarly influence as citing works accumulate over time.
Focus0/100
The focus axis scores 0, though this likely reflects the tiny indexed sample (4 works) rather than genuine scattering. The topic distribution actually clusters: Software Engineering Research (4), Testing and Debugging Techniques (3), and SE Techniques and Practices (2), with single counts for Topic Modeling and logic/type systems. The field split (0.5/0.5) falls between two adjacent areas—LLMs for code and LLM-based code generation—suggesting a coherent LLM-for-software-engineering agenda.
Collaboration39/100
The collaboration axis scores 39, computed from 6 distinct coauthors in the indexed window. The frequent-coauthor list—Jayanth Srinivasa, Milos Gligoric, Aditya Thimmaiah, Linghan Zhong, Lara Marinov, and Jiyang Zhang—indicates a compact, recurring author team rather than large rotating consortia, with all three reviewed 2026 papers written within this stable circle.
Activity47/100
The activity axis scores 47. OpenAlex counts 4 works in the recent three-year window (countsByYear: 1 in 2024, 3 in 2026), matching the recent3yWorks figure of 4. The 2026 cluster—three arXiv preprints on LLMs and code—marks a visible burst of output after the single 2024 record, indicating a currently active agenda with a short observed history.
Reproducibility78/100
The reproducibility axis scores 78, the strongest signal in this profile, derived from a reproducibility mean of 78 across 2 assessed papers. Two of the three catalog papers—the exception-retrofitting study and the PrEx exit-code benchmark—carry high reproducibility readiness, while CROCODIL is rated medium. Consistent evaluation across 5 LLMs and public benchmark construction support this standing.
Research summary
Junyi Jessy Li is a researcher at The University of Texas at Austin whose indexed work centers on applying large language models to software engineering: code generation, code editing, and evaluation against formal program semantics. The OpenAlex snapshot is small (4 works; h-index and i10-index both 0), so output- and impact-level conclusions are premature, but the topical signal is coherent: Software Engineering Research (4 topic hits), Testing and Debugging Techniques (3), and Software Engineering Techniques and Practices (2), with the field split evenly between LLMs for code and LLM-based code generation. Three 2026 arXiv preprints reviewed by Evolve define the current agenda. One retrofits existing code with exception handling derived from tests, reporting consistent gains across 5 LLMs while flagging that ~25% of test-passing code deviates from intent. A second, CROCODIL, targets excessive edits when LLMs modify others' code, halving Olmo3 7B's edit volume via a GRPO reward. The third introduces the PrEx benchmark, pairing LLMs with operational semantics to predict program exit codes and showing open-source models answer from priors rather than supplied semantics. Reproducibility is the profile's strongest dimension (score 78, a mean across 2 assessed papers). Collaboration is compact: 6 distinct coauthors in a recurring group including Milos Gligoric and Jayanth Srinivasa, with middle-to-last authorship positions and no corresponding-author credits in the catalog.
Collaboration pattern
Collaboration runs through a compact recurring team: Jayanth Srinivasa, Milos Gligoric, Aditya Thimmaiah, Linghan Zhong, Lara Marinov, and Jiyang Zhang, with 6 distinct coauthors recorded and a collaboration score of 39. Across the three 2026 arXiv papers the researcher holds middle or last authorship positions, none corresponding, consistent with a senior-collaborator role on LLM-for-code projects rather than solo work.
Research keywords
Highlights
Retrofitting exception handling into existing code with LLMs
CROCODIL curbs LLMs' excessive edits on others' code
PrEx benchmark tests whether LLMs actually use given semantics
Timeline
- 2024Earliest record in the current OpenAlex snapshot: 1 work indexed for this year.
- 2026Preprint on retrofitting code with LLM-generated exception handling from tests (arXiv; middle authorship, high reproducibility readiness).
- 2026CROCODIL preprint on cross-model code editing; a GRPO reward halves Olmo3 7B edits while preserving success (arXiv; last authorship).
- 2026PrEx benchmark paper predicting program exit codes with LLMs and programming-language semantics (arXiv; high reproducibility readiness).
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 indicators come from a single OpenAlex record only: h-index and i10-index are 0 and just 4 works are indexed, so the output and impact axes are unscored and citation-based measures likely understate reality.
- This profile is name-resolved from public bibliographic data (OpenAlex identity; ORCID 0000-0002-2550-5262); verify the record matches the intended person, since the indexed topic mix is software-engineering-focused.
- All three catalog papers are 2026 arXiv preprints; peer-review status and citation history are not yet reflected in this profile.
- The reproducibility score (mean 78) covers 2 assessed papers; the third paper, CROCODIL, is rated medium, and its reported gains lack statistical testing per the Evolve review.