Md Tahmid Rahman Laskar

—

Unclaimed · AI-structured profileORCID OpenAlex

NLP/ML researcher diagnosing LLM agent evaluation and MoE+LoRA fine-tuning, with cultural multimodal benchmarking work

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

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—

No publication-volume metrics are available yet, so output is unscored. This axis would measure total papers, yearly production, and authorship position — none of which can be derived from the three cataloged 2026 preprints alone.

Impact—

Impact is unscored because citation and influence data are unavailable in the input. This axis would measure citation counts, h-index, and downstream uptake — signals that cannot be inferred for 2026 preprints from the provided record.

Focus0/100

The focus score is 0, derived from a field distribution split exactly 50/50 between natural language processing (LLM agent evaluation) and machine learning (LLM fine-tuning / parameter-efficient adaptation). A third paper on South Asian multimodal benchmarking sits outside both clusters, reinforcing breadth over concentration. The record so far describes a deliberately dual-track early-stage career rather than a single specialized line of inquiry.

Collaboration66/100

Collaboration scores 66 on a 0–100 scale, based on 27 distinct coauthors across the cataloged papers. Twelve recurring collaborators are visible, including Xue-Yong Fu, Gundeep Singh, Karol Chang, Kevin Sanders, Shi Zong, and Tania Habib, alongside Julien Bouvier Tremblay, Shayna Gardiner, Harsh Saini, Matthias Lee, Elena Khasanova, and Quinten McNamara — a pattern consistent with team-based, lab- or industry-style group authorship.

Activity—

Activity is unscored for lack of longitudinal data. This axis would measure recent publication cadence and sustained research tempo; the record shows three papers all dated 2026, too narrow a window to score.

Reproducibility38/100

The reproducibility score is 38, the mean over the 2 papers with repro-readiness ratings. The first-author agent-evaluation paper rates low, with Evolve citing proprietary training data, a tiny evaluation set, and no significance tests; SpawnLoRA and MemeCULT-1K both rate medium, with confounded baselines and annotator-overlap issues respectively. Open artifacts and statistical rigor are the clearest levers for improvement.

Research summary

Md Tahmid Rahman Laskar (ORCID 0009-0001-7880-299X) is an early-career NLP/ML researcher whose identity is name-resolved from public bibliographic data; no institutional affiliation or country is present in the input record. The Evolve catalog holds three 2026 arXiv preprints, split evenly between two fields: LLM agent evaluation (50%) and LLM fine-tuning / parameter-efficient adaptation (50%). As first author of 'When Better Turns Do Not Make Better Agents', he shows supervised fine-tuning lifts next-turn metrics while achieving 0/77 holistic workflow success — a controlled design Evolve credits, while flagging proprietary data, a tiny evaluation set, and no significance tests. As last author of 'Routing Is Not Enough', he diagnoses intra-adapter subspace contention in MoE+LoRA fine-tuning and proposes the gated sub-adapter method SpawnLoRA, with suggestive but baseline-confounded evidence. He also contributes as a middle author to MemeCULT-1K, a 1,000-meme South Asian cultural-humor benchmark evaluating 13 vision-language models. Available bibliometrics report 27 distinct coauthors, a collaboration score of 66, a focus score of 0 reflecting the even two-field split, and a reproducibility score of 38 averaged over 2 rated papers; output, impact, and activity scores are not yet available.

Collaboration pattern

The record shows 27 distinct coauthors across three 2026 preprints, with 12 recurring collaborators: Xue-Yong Fu, Gundeep Singh, Karol Chang, Kevin Sanders, Shi Zong, Tania Habib, Julien Bouvier Tremblay, Shayna Gardiner, Harsh Saini, Matthias Lee, Elena Khasanova, and Quinten McNamara. Authorship positions span first, middle, and last, suggesting movement between leading evaluation studies and supporting engineering-heavy fine-tuning work within a stable team.

Research keywords

LLM agentsnext-turn evaluationworkflow successsupervised fine-tuningMixture-of-ExpertsLoRAparameter-efficient fine-tuningnegative transfermultimodal benchmarksvision-language modelscultural knowledgememe understanding

Highlights

Next-turn metrics improve under SFT, but workflow success does not
First-author study shows supervised fine-tuning lifts next-turn measures yet yields 0/77 holistic workflow success; Evolve credits the controlled design while flagging proprietary data, a tiny evaluation set, and missing significance tests.
SpawnLoRA targets adapter contention that MoE routing cannot resolve
Last-author work diagnoses intra-adapter subspace contention in MoE+LoRA fine-tuning and proposes the gated sub-adapter SpawnLoRA; Evolve finds the small-scale evidence suggestive but the baselines confounded.
MemeCULT-1K benchmarks South Asian cultural and humor understanding
Contributing to a 1,000-meme benchmark, the team shows cultural context improves explanations from 13 vision-language models; Evolve notes an annotator-overlap confound and some non-significant cells temper the headline claim.

Timeline

2026First-author arXiv preprint 'When Better Turns Do Not Make Better Agents' appears, reporting that supervised fine-tuning lifts next-turn metrics while holistic workflow success stays at 0/77.
2026Last-author preprint 'Routing Is Not Enough' attributes MoE+LoRA interference to intra-adapter subspace contention and proposes SpawnLoRA.
2026Contributes as a middle author to MemeCULT-1K, a 1,000-meme South Asian cultural-humor benchmark evaluating 13 vision-language models.

Papers in the Evolve catalog

Frequent co-authors

Xue-Yong Fu×1 Gundeep Singh×1 Karol Chang×1 Kevin Sanders×1 Shi Zong×1 Tania Habib×1 Julien Bouvier Tremblay×1 Shayna Gardiner×1 Harsh Saini×1 Matthias Lee×1 Elena Khasanova×1 Quinten McNamara×1

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 cataloged papers are 2026 arXiv preprints; no peer-reviewed venue information is included in the input data.
Institution and country are null in the source record; identity is name-resolved from public bibliographic data via ORCID 0009-0001-7880-299X.
Output, impact, and activity scores are missing; bibliometric depth is limited to collaboration, focus, and reproducibility signals.
The reproducibility mean of 38 is computed over only 2 rated papers and should not be read as a stable estimate.