Abdullah Al Shafi

Khulna University of Engineering and Technology · Bangladesh

Unclaimed · AI-structured profileOpenAlex

Early-career KUET researcher working across low-resource Bangla NLP, multimodal QA, and ophthalmology AI

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

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 provides no verified total publication count, and the h-index and i10-index are both 0. This axis would measure cumulative output volume and multi-year publishing consistency once fuller production records accumulate.

Impact—

Not yet scored: no citation data is available and the h-index and i10-index stand at 0. This axis would measure citation uptake, citing works, and scholarly influence once the profile builds a measurable citation footprint.

Focus0/100

The focus score is 0 because output is split evenly across three loosely related fields: multimodal question answering, low-resource Bangla NLP, and ophthalmology AI each hold 0.33 of the field distribution. Topics mirror the spread — topic modeling, advanced neural network applications, and NLP techniques at 4 mentions each, beside medical image segmentation (2) and sentiment analysis (2). The pattern reads as exploratory breadth rather than a consolidated specialty in these early records.

Collaboration46/100

With a collaboration score of 46, this is a team-based but compact network: 9 distinct coauthors appear across 13 recent three-year works. Recurring partners include Swapnil Kundu Argha, Rowzatul Zannat, Shoumik Barman Polok, Abdul Muntakim, Jannatul Ferdousi, M. A. Moyeen, Nishat Sadaf Lira, Abrar Hasan, and Kazi Saeed Alam. Authorship roles vary — first author on two catalog papers, middle author on one, corresponding author once — an emerging group pattern around KUET.

Activity77/100

Activity scores 77, the strongest axis: OpenAlex counts 13 works in the recent three-year window, jumping from 1 in 2025 to 12 in 2026, and all three Evolve-reviewed papers are 2026 arXiv preprints. This concentrated burst indicates a rapid entry into publishing, though the single-year clustering means sustainability beyond 2026 cannot yet be judged from the available counts-by-year data.

Reproducibility43/100

The reproducibility mean is 43 across 3 assessed papers. Two catalog papers — the Bangla sentence-function corpus study and the glaucoma detection study — are rated medium readiness, while the multimodal QA synthesis is rated low, as it re-analyzes published results without new experiments. Evolve reviews flag missing transformer baselines, an absent fine-tuned CNN comparison, and no significance tests as the main gaps limiting replication confidence.

Research summary

Abdullah Al Shafi is an early-stage researcher at Khulna University of Engineering and Technology (KUET), Bangladesh, whose 2026 output spans three distinct areas: low-resource Bangla language technology, multimodal question answering, and medical image analysis for ophthalmology. OpenAlex records 13 works in the recent three-year window — 12 in 2026 and 1 in 2025 — while the h-index and i10-index both stand at 0, consistent with a citation record just beginning. Three arXiv preprints are in the Evolve catalog: a first-author, corresponding-author synthesis of multimodal QA arguing that unified text-centric PLM pipelines prevail; a first-author glaucoma detection study pairing CBAM-attention InceptionV3 features with heterogeneous ensembles (0.90 accuracy on EDC); and a contribution to a 10k-sentence Bangla sentence-function corpus (κ=0.82) that reaches 0.95 F1 with ML ensembles. Topic data concentrates on topic modeling, advanced neural network applications, and NLP techniques. Evolve reviews flag recurring methodological gaps — missing transformer or fine-tuned CNN baselines and no significance tests — with a reproducibility mean of 43 across three papers. Collaboration is moderate, with 9 distinct coauthors in the recent window.

Collaboration pattern

Collaboration is moderate (score 46): 9 distinct coauthors across 13 recent three-year works, forming a compact, KUET-centered circle. Frequent partners include Swapnil Kundu Argha, Rowzatul Zannat, Shoumik Barman Polok, Abdul Muntakim, Jannatul Ferdousi, M. A. Moyeen, Nishat Sadaf Lira, Abrar Hasan, and Kazi Saeed Alam. Roles rotate between first author (two catalog papers), corresponding author (one), and middle author (one), consistent with an early-stage group publishing together.

Research keywords

Bangla NLPlow-resource language technologysentence function classificationcorpus annotationmultimodal question answeringcross-modal reasoningtopic modelingsentiment analysisglaucoma detectionretinal fundus imagingattention mechanisms (CBAM)ensemble learning

Highlights

10k-sentence Bangla sentence-function corpus with strong inter-annotator agreement
A middle-authorship contribution delivering a 10k-sentence Bangla sentence function corpus with Fleiss' κ=0.82, reaching 0.95 F1 using ML ensembles over TF-IDF and Word2Vec features; Evolve notes the absence of transformer baselines and significance tests. Reproducibility readiness: medium.
First-author synthesis of multimodal QA argues for unified text-centric pipelines
As first and corresponding author, this arXiv paper compares MAE, Solar, and UniMMQA for multimodal question answering and argues that unified text-centric PLM pipelines prevail. Evolve classifies it as a synthesis of published results with no new experiments, rated low reproducibility readiness.
Attention-enhanced ensemble learning reaches 0.90 accuracy for glaucoma detection
First-author work combining CBAM-attention InceptionV3 deep features with heterogeneous ensemble learning and SMOTE Tomek Links resampling reports 0.90 accuracy on EDC retinal fundus data. Evolve finds the pipeline sound but the gains small and untested, with a fine-tuned CNN baseline missing. Reproducibility readiness: medium.

Timeline

2025Earliest visible activity: OpenAlex counts-by-year logs 1 work in 2025, marking the start of the publishing record.
2026Publishes a first-author, corresponding-author arXiv synthesis on the evolution of multimodal question answering, comparing MAE, Solar, and UniMMQA.
2026First-author arXiv paper on glaucoma detection combines CBAM-attention InceptionV3 features with heterogeneous ensemble learning, reporting 0.90 accuracy on EDC.
2026Contributes as middle author to a 10k-sentence Bangla sentence-function corpus paper; OpenAlex logs 12 works in 2026, the profile's most active year to date.

Papers in the Evolve catalog

2026-09-12
Bangla Sentence Function Classification: Corpus Development, Model Benchmarking, and Interpretability
2026-09-08first authorcorresponding
Evolution of Multimodal Question Answering: From Modality-Adaptive Extraction to Unified Language Representation
2026-09-06first author
Attention-Enhanced Deep Features with Heterogeneous Ensemble Learning for Glaucoma Detection

Frequent co-authors

Swapnil Kundu Argha×2 Rowzatul Zannat×1 Shoumik Barman Polok×1 Abdul Muntakim×1 Jannatul Ferdousi×1 M. A. Moyeen×1 Nishat Sadaf Lira×1 Abrar Hasan×1 Kazi Saeed Alam×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.

Bibliometrics are derived from OpenAlex only; no ORCID is linked, so this profile is name-resolved from public bibliographic data and may merge or split namesakes.
The h-index and i10-index are 0 in the snapshot, and the output and impact axes are left unscored for lack of citation data.
All three catalog papers are 2026 arXiv preprints; peer-review status and final venue information are not captured.
Reproducibility assessments cover only the 3 papers reviewed by Evolve and may not reflect the full body of work.