CUHK researcher scaling LLM reasoning and agentic coding — from self-correction data to synthesized RL environments
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
26
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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 is unscored: the snapshot lacks a verified, de-duplicated total publication count, so research volume cannot be computed. This axis would measure total works, annual output, and first- or corresponding-author share across the full publication record.
Impact—
Impact is unscored: citation totals and venue-level influence are not available in this snapshot, so no estimate is offered. The axis would measure citation counts and velocity, h-index trajectory, and the uptake of this work by other researchers.
Focus0/100
The focus score is 0, which here reflects breadth rather than inactivity: the field distribution splits evenly, 0.5 to LLM reasoning NLP and 0.5 to LLM coding agents. Topic data reinforces the spread — Topic Modeling (84), NLP Techniques (72), Multimodal Machine Learning Applications (33), and Domain Adaptation and Few-Shot Learning (25) — so no single field dominates the recent record.
Collaboration65/100
The collaboration score is 65. OpenAlex records 25 distinct coauthors over the recent three-year window alongside 49 works in that period, and the input lists 12 recurring collaborators including Qirui Chen, Renjie Pi, Jiahui Gao, and Lei Li. Both catalogued 2026 papers are team efforts, with Kong positioned as a middle author on CodeMidas and last author on Reflective Recovery, neither as corresponding author — a supportive, group-embedded pattern.
Activity100/100
Activity scores 100, the maximum. Annual work counts in the snapshot jump from 5 in 2021 to 21 in 2022, peak at 28 in 2023, and stay high at 24 in 2024 and 18 in 2025, with 7 recorded in 2026 to date. The last three years alone account for 49 works, an unusually dense and sustained publication cadence.
Reproducibility38/100
Reproducibility averages 38 across the 2 papers Evolve has assessed. CodeMidas is rated low readiness: Evolve judges its reported gains on five benchmarks for MiMo-V2.5 plausible but based on single runs with no significance tests. Reflective Recovery is rated medium, with consistent but modest improvements over rejection-sampling fine-tuning and unresolved gaps in the scaling-collapse evidence. The pattern is typical of fast-moving preprint work.
Research summary
Lingpeng Kong is a researcher at the Chinese University of Hong Kong whose recent work concentrates on large language model reasoning and LLM coding agents; the underlying field distribution splits exactly 50/50 between the two. The OpenAlex record shows an h-index of 26 and an i10-index of 40, with a sharp acceleration in output: annual work counts rise from 5 in 2021 to 21 in 2022, peak at 28 in 2023, and stay high at 24 (2024) and 18 (2025), totalling 49 works in the last three years. Topic fingerprints are led by Topic Modeling (84), NLP Techniques (72), and Multimodal Machine Learning Applications (33). Two 2026 arXiv preprints reviewed by Evolve define the current agenda. CodeMidas claims source code alone can yield 5,545 reinforcement learning environments that lift MiMo-V2.5 on five benchmarks, though gains are single-run and untested for significance. Reflective Recovery converts failed reasoning traces into recovery training data, improving consistently but modestly over rejection-sampling fine-tuning. Both are team efforts, with Kong as a middle author on the first and last author on the second, neither corresponding. Evolve's reproducibility read is cautious — 38 average readiness across two papers.
Collaboration pattern
Kong works in sizeable teams: the recent three-year window shows 25 distinct coauthors across 49 works, with 12 recurring collaborators — among them Qirui Chen, Renjie Pi, Jiahui Gao, Bowen Ye, Lei Li, Shicheng Li, Zihao Yue, Linghao Zhang, Hanglong Lv, Yuanxin Liu, Wenhan Ma, and Hao Tian. On both catalogued 2026 papers Kong is a non-corresponding author, in middle position on CodeMidas and last position on Reflective Recovery, consistent with a senior-group-member collaboration style (collaboration score 65).
Research keywords
Highlights
Scaling agentic RL environments from code itself
Turning failed reasoning traces into training data
Timeline
- 2011Earliest publication year recorded in the OpenAlex snapshot, with 1 work.
- 20215 works recorded, marking the start of a steep multi-year ramp in output.
- 2022Output jumps to 21 works, more than four times the 2021 count.
- 2023Peak annual output at 28 works, the highest in the record.
- 2026Two arXiv preprints enter the Evolve catalog — CodeMidas and Reflective Recovery; the snapshot logs 7 works in total for the year so far.
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 figures come from a single OpenAlex-based snapshot; output and impact dimensions are missing from the data and are not scored.
- Paper-level evaluation covers only the two 2026 arXiv preprints reviewed by Evolve; other publications are not individually assessed.
- The reproducibility score rests on just 2 assessed papers and may not reflect the wider body of work.
- Identity is name-resolved from public bibliographic data (OpenAlex, ORCID 0000-0002-9033-2724), with the affiliation as given in the source snapshot.