NLP researcher at BASF (Canada) working on on-device function calling and cross-lingual retrieval for chemical patents
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-09) 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 provided record contains only two 2026 arXiv papers, with no totals for publication volume, venues, or yearly productivity. This axis would measure cumulative output and its growth over time.
Impact—
Not yet scored: no citation counts, h-index, or adoption signals exist in the provided data. This axis would measure how widely the two reviewed papers are cited and used by other researchers.
Focus0/100
Scores 0 on focus, the lowest concentration: the field distribution splits exactly 0.5/0.5 between natural language processing for on-device machine learning and NLP/information retrieval. Both are applied, adjacent areas — mobile function calling and cross-lingual patent retrieval — so the split reflects two parallel applied tracks rather than scattered interests.
Collaboration41/100
Scores 41 out of 100, reflecting a compact collaboration footprint: 7 distinct coauthors in total, drawn from a recurring team that includes Ali Rezagholizadeh and Hamidreza Mahyar. Soheila Samiee takes the last author position on both papers without corresponding status, a pattern consistent with senior supervision inside one small, stable group.
Activity—
Not yet scored: the data covers only 2026, with no yearly publication cadence, prior-year history, or recency measures. This axis would measure publishing rhythm and sustained research activity over time.
Reproducibility100/100
Scores 100, the maximum, computed across both reviewed papers (reproMean 100 over reproPapers 2), each rated high for repro-readiness. The ChemCLIR-Bench benchmark and the open FunctionGemma work are built for reuse; the remaining gaps are evaluation scope — synthetic-only tests and thin baseline coverage — rather than artifact availability.
Research summary
Soheila Samiee is a researcher at BASF (Canada) whose cataloged work sits at the intersection of applied natural language processing and information retrieval. Two 2026 arXiv papers place her in the last author position. The first extends FunctionGemma into a practical on-device mobile function-calling system: a 270M-parameter model whose end-to-end accuracy rises from 29.3% to 76.5% using 9.5K synthetic examples — work Evolve judges credible and open, though evaluated only on synthetic data without significance testing. The second, ChemCLIR-Bench, introduces a five-language benchmark for cross-lingual information retrieval over multilingual chemical patents, showing that dense retrievers degrade sharply across languages; Evolve credits the analysis while flagging a tiny query set and missing machine-translation and sparse baselines. Bibliometrically, her cataloged output splits evenly (0.5/0.5) between NLP for on-device machine learning and NLP/IR, and seven distinct coauthors form a recurring core team, yielding a collaboration score of 41. Both papers earn high repro-readiness ratings (reproMean 100 across 2 papers), indicating deliberate attention to reusable artifacts. Output, impact, and activity metrics are not yet available, so this early-stage profile rests on a narrow evidence base and is name-resolved from public bibliographic data.
Collaboration pattern
Collaboration is compact and stable: 7 distinct coauthors, with Ali Rezagholizadeh, Mahdi Astaraki, Mohammad Khodadad, Reza Namazi, Mohammad Arshi Saloot, Amir Reza Behzad Moghadam, and Hamidreza Mahyar recurring across the catalog. Both papers place Soheila Samiee in the last author position without corresponding-author status, consistent with senior supervision inside one core team rather than a cross-institution network; the collaboration dimension scores 41 on a 0-100 scale.
Research keywords
Highlights
Lifting on-device function calling from 29.3% to 76.5%
ChemCLIR-Bench: five-language chemical-patent retrieval benchmark
High reproducibility across both reviewed papers
Both 2026 papers earn a high repro-readiness rating, giving a reproducibility mean of 100 across 2 papers — the reusable ChemCLIR-Bench benchmark and the open function-calling work are positioned for third-party reuse.
Timeline
- 2026Published on arXiv 'Extending FunctionGemma for Practical On-Device Mobile Function Calling', in the last author position.
- 2026Published on arXiv 'ChemCLIR-Bench: Benchmarking Cross-Lingual Information Retrieval in Multilingual Chemical Patents', again as last author.
- 2026Both papers received high repro-readiness ratings, yielding a reproducibility mean of 100 across 2 papers in the Evolve radar.
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.
- Dimension scores are computed from the platform's indexed bibliometric snapshot only; citation-based output and impact data are not yet available.
- The evidence base is two 2026 arXiv papers reviewed by Evolve; earlier publications, if any, fall outside this profile's scope.
- No ORCID or country field was available, so identity is name-resolved from public bibliographic data.
- Reproducibility reflects Evolve's review ratings (mean 100 over 2 papers), not an external replication study.