Speech/NLP researcher spanning efficient zero-shot SpeechLLMs and consumer-side audits of AI-generated recommendations.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-20) 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 or output metrics are available in the current snapshot, so this axis is unscored. It would measure total publication count, venue spread, and authorship-role distribution across the researcher's full record.
Impact—
Impact cannot be computed: the snapshot contains no citation counts, h-index, or influence data. This axis would measure citations, field-normalized influence, and uptake of the two reviewed papers once such data is available.
Focus0/100
Focus scores 0/100, reflecting an almost perfectly even split in the field distribution: 50% speech and natural language processing versus 50% computational social science / algorithm auditing (LLMs) across the two reviewed papers. Rather than diffusion, this reads as a deliberate dual-track agenda; distinguishing genuine breadth from scattered focus would require a larger paper sample than the current snapshot provides.
Collaboration50/100
Collaboration scores 50/100, computed from 11 distinct coauthors in the bibliometric snapshot. The frequent-coauthor list spans both of the researcher's fields — including Lucas G. Uberti-Bona Marin, Thales Bertaglia, Giovanni Astante, Bram Rijsbosch, Gijs van Dijck, Anikó Hannák, Konrad Kollnig, Abderrahmane Issam, Yusuf Can Semerci, and Jan Scholtes — though the snapshot does not map individuals to specific papers. The mid-range score points to steady, mid-sized team science rather than large consortia.
Activity—
No year-by-year productivity data is present, so recent activity is unscored. This axis would measure publication cadence and consistency over time; both reviewed papers carry 2026 dates, hinting at current activity, but that is insufficient to score.
Reproducibility78/100
Reproducibility is the strongest scored axis at 78/100, the mean across the 2 papers carrying reproducibility assessments. Evolve rates the DTW-aligned CIF SpeechLLM paper "high", while noting its headline gains rest on single-seed runs; the AI-recommendation audit paper is rated "medium", with the reasons not further specified in the snapshot. Given only two assessed papers, treat this as an early signal rather than a settled record.
Research summary
Gerasimos Spanakis is a researcher working across two strands that rarely meet: speech-and-language systems engineering and the algorithmic auditing of large language models. Evolve's catalog carries two 2026 arXiv papers. The first, "Align, Integrate, and Fire," uses Dynamic Time Warping to align Continuous Integrate-and-Fire (CIF) token representations with single-layer knowledge distillation to build efficient zero-shot SpeechLLMs for recognition and translation; Evolve judges the gains over SpeechMapper credible while cautioning that headline claims rest on single-seed runs, and rates reproducibility readiness high. The second, "If I Had to Buy Just ONE: Galaxy S26 Ultra," audits 1,536 AI-generated product recommendations and finds that AI shopping advice is unstable and that API-based audits miss what consumers actually see, with causality and generalizability limits; its reproducibility readiness is medium. The dual track shows up quantitatively: the field distribution is an even 50/50 split between speech/NLP and computational social science/algorithm auditing, which is why the focus score is 0. Reproducibility is the strongest signal at 78 across two assessed papers, and collaboration sits at a mid-range 50 with 11 distinct coauthors recorded. Output, impact, and activity metrics are absent from the snapshot and remain unscored. On both papers Spanakis appears as last or middle author, never corresponding — a senior-contributor pattern within small, focused teams rather than large consortia.
Collaboration pattern
The snapshot records 11 distinct coauthors, and the frequent-collaborator list includes Lucas G. Uberti-Bona Marin, Thales Bertaglia, Giovanni Astante, Bram Rijsbosch, Gijs van Dijck, Anikó Hannák, Konrad Kollnig, Abderrahmane Issam, Yusuf Can Semerci, and Jan Scholtes. The mix spans speech/NLP and algorithm-auditing communities, consistent with the 50/50 field split; on the two reviewed papers Spanakis appears as a middle or last author rather than corresponding author, suggesting a senior-contributor role within small, focused teams.
Research keywords
Highlights
Efficient zero-shot SpeechLLMs via DTW-aligned CIF and single-layer distillation
Auditing 1,536 AI-generated product recommendations from the consumer's side
A deliberate two-track profile: speech systems and LLM auditing in parallel
The snapshot's field distribution is an exact 50/50 split between speech and natural language processing and computational social science / algorithm auditing (LLMs) — the direct source of the 0 focus score. Both strands produced arXiv work in 2026, suggesting two parallel research tracks rather than a transition from one field to another.
Timeline
- 2026Posted "Align, Integrate, and Fire" to arXiv, introducing DTW-aligned CIF and single-layer knowledge distillation for efficient zero-shot SpeechLLMs, as last author.
- 2026Co-authored the consumer-side audit of 1,536 AI-generated product recommendations, a middle-author contribution to the algorithm-auditing track.
- 2026Published across both tracked fields in the same year — speech/NLP engineering and computational social science / algorithm auditing — matching the snapshot's 50/50 field split.
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.
- Profile is name-resolved from public bibliographic data; institution, country, and ORCID are absent from the snapshot, so identity confidence is moderate.
- Only two reviewed papers underpin this profile; the field distribution and reproducibility mean may shift as the catalog grows.
- Output, impact, and activity axes lack underlying data entirely and are left unscored rather than estimated.
- The frequent-coauthor list includes the researcher's own name, an apparent data artifact; coauthor-to-paper mappings are not recorded.