Amazon researcher on RL for LLM reasoning and agents; earlier record spans IR, semantic web, and topic modeling.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
21
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—
Not yet scored: the snapshot lacks publication-volume data for this axis. Output would measure cumulative works, annual productivity, and first-author share; the raw record's yearly counts (1–16 works per year, 2012–2026) await aggregation into a score.
Impact—
Not yet scored: no citation-impact dimension is computed in this snapshot. Impact would measure field-normalized citation rates and highly-cited works; the raw record lists h-index 21 and i10-index 32, but a normalized impact score has not been derived.
Focus0/100
Scores 0/100, the lowest reading on this radar. The raw field distribution is an even 50/50 split between two machine-learning directions — efficient reasoning in LLMs and LLM agents/reinforcement learning — atop an earlier topical base spanning five clusters (topic modeling 57, semantic web 33, IR 32, NLP 29, recommender systems 12). This reads as a deliberate research pivot in transition rather than drift; treat the score as dispersion, not inactivity.
Collaboration41/100
Scores 41/100. The recent record shows 7 distinct co-authors in the last three years, and both Evolve-reviewed 2026 papers place him as a middle author without the corresponding role — consistent with contribution inside a larger Amazon team. Recurring collaborators include Ruochen Jiao, Zhenyu Shi, Priyanka Nigam, Xinlu Zhang, Ying-Chun Lin, Zhihan Zhang, and Xi Chen, suggesting a stable core group rather than a wide external network.
Activity84/100
Scores 84/100, the strongest axis on this radar. The raw record shows 17 works in the last three years and continuous output from 2012 to 2026, with annual counts of 16 (2022), 10 (2023), 8 (2024), 3 (2025), and 6 so far in 2026 — a mid-decade peak followed by steady ongoing production, including the two 2026 arXiv preprints reviewed by Evolve.
Reproducibility55/100
Scores 55/100 (reproMean 55 across 2 assessed papers). Both Evolve-reviewed preprints are rated medium for reproducibility: CounterRoute relies on single-seed, author-run baselines, and CHART is well-ablated but single-run and single-model, with no code released. The pattern suggests solid experimental effort, with gaps in statistical robustness and artifact availability that multi-seed runs and public code would close.
Research summary
Besnik Fetahu is a researcher at Amazon (United States) whose public OpenAlex record spans 2012–2026, with an h-index of 21 and an i10-index of 32. His bibliometric profile is rooted in information access and text mining — topic modeling (57 works), semantic web and ontologies (33), information retrieval and search behavior (32), NLP techniques (29), and recommender systems (12) — and has shifted toward machine learning on large language models, split evenly between efficient LLM reasoning and LLM agents/reinforcement learning. Activity is high (score 84): 17 works in the last three years, with annual counts peaking at 16 in 2022 and continuing through 2026. Evolve has reviewed two 2026 arXiv preprints where he is a middle author: CounterRoute, which uses hierarchical counterfactual credit assignment in a GRPO pipeline so Qwen3 models self-route between thinking and direct answers, reportedly cutting tokens 41–51% while lifting accuracy; and CHART, a harness-rotation curriculum that addresses harness-local RL behavior in search agents. Both are rated medium for reproducibility, with single-seed or single-run baselines and limited code release noted; the reproducibility score is 55 across 2 assessed papers. Collaboration scores 41, with 7 distinct recent co-authors including Ruochen Jiao, Zhenyu Shi, and Xi Chen. Output and impact dimension scores are not yet computed in this snapshot.
Collaboration pattern
The recent record shows 7 distinct co-authors over the last three years, and both 2026 preprints list him as a middle, non-corresponding author — the signature of contribution within a larger industrial research team at Amazon. Repeated collaborator names (Ruochen Jiao, Zhenyu Shi, Priyanka Nigam, Xinlu Zhang, Ying-Chun Lin, Zhihan Zhang, Xi Chen) point to a stable internal group; no external-academic co-authorship pattern is visible in this snapshot.
Research keywords
Highlights
Self-routed reasoning cuts LLM thinking tokens by 41–51%
Harness-rotation curriculum targets harness-local RL behavior
Timeline
- 2012Earliest year in the OpenAlex yearly counts, with 1 recorded work.
- 2013Early peak of 14 recorded works, the second-highest annual count in the record.
- 2022Highest annual output in the record: 16 works, opening a sustained recent stretch.
- 20248 works; the last three years together total 17 works, backing an activity score of 84.
- 2026Two Evolve-reviewed arXiv preprints on RL for LLM reasoning and agents — CounterRoute and CHART — listing him as middle author.
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.
- All bibliometrics (h-index, yearly counts, topics, co-author counts) come from the OpenAlex record; no other citation source is included.
- Output and impact dimension scores are not computed in this snapshot and are left unscored rather than estimated.
- Evolve review coverage is narrow: two 2026 arXiv preprints with middle authorship; earlier career work is not reviewed here.
- This profile is name-resolved from public bibliographic data only; biographical details beyond the listed affiliation are not verified.