Document AI / NLP researcher working on cost-efficient LLM methods for document QA and page stream segmentation
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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 score yet: the available records cover only two reviewed 2026 arXiv papers, too thin to measure publication volume, venue spread, or yearly productivity. This axis would track total and per-year research output once a fuller publication record is linked.
Impact—
No score yet: no citation counts, h-index, or uptake data are available for this profile. This axis would measure citations and influence on document-AI practice once such bibliometric signals can be reliably attached to this name-resolved record.
Focus0/100
The focus axis scores 0 on the 0-100 scale. The raw field distribution behind it is an even 50/50 split between two adjacent document-AI/NLP subfields — page stream segmentation and multimodal document AI — drawn from only two catalogued 2026 papers. The zero should be read cautiously: the sample is very small and the two strands are thematically close (cost-efficient LLM processing of documents), so the score reflects sparse data as much as any demonstrated scattering of interests.
Collaboration36/100
Collaboration scores 36/100, computed from a small footprint: 5 distinct coauthors across the reviewed record, with Nikhil Reddy Pottanigari, Sepideh Kharaghani, Saverio Vadacchino, and Ying Zhang recurring. Both 2026 papers list this researcher in a middle author position without corresponding-author credit, consistent with contributing-member roles in small teams rather than solo or leading-author work.
Activity—
No score yet: the input lacks publication dates beyond the two 2026 papers, so cadence and recency of work cannot be assessed. This axis would measure publishing rhythm and ongoing research engagement across a longer time window.
Reproducibility55/100
Reproducibility scores 55/100, a mean taken over 2 papers with reproducibility assessment, both rated medium readiness. The evidence base is single-run on each paper: Evolve's verdicts note no significance testing for the document-QA representation comparison and single-run evidence lacking significance checks for MSBD, so the middling score tracks honest, controlled setups whose claims are not yet statistically hardened.
Research summary
Alejandro Posada is an early-stage researcher in natural language processing and document AI, profiled here from public bibliographic records and two 2026 arXiv papers reviewed in the Evolve catalog. Both papers sit in adjacent document-AI subfields — multimodal document question answering and LLM-based page stream segmentation — giving an even 50/50 field split in the reviewed work, and both target the same underlying question: how to obtain accurate document understanding from large language models at lower inference cost. 'Text, Pixels, or Both?' compares text, image, and combined input representations for document QA within a 50-page cap, finding accuracy gains from page images that come with growing latency cost; Evolve's verdict calls the work controlled and honest but flags missing significance testing. 'Splitting Documents at Lower Cost' proposes MSBD, which batches page-boundary decisions into a single LLM call with accuracy held at a model-specific knee; the supporting evidence is again single-run. Authorship is a middle position on both papers without corresponding-author credit, and the recorded collaboration footprint is small — five distinct coauthors, with Nikhil Reddy Pottanigari, Sepideh Kharaghani, Saverio Vadacchino, and Ying Zhang recurring. Radar scores place reproducibility at 55/100 (medium readiness on both papers), collaboration at 36/100, and focus at 0; output, impact, and activity axes lack sufficient data and remain unscored.
Collaboration pattern
The recorded collaboration footprint is compact: 5 distinct coauthors, yielding a collaboration score of 36/100. Nikhil Reddy Pottanigari, Sepideh Kharaghani, Saverio Vadacchino, and Ying Zhang recur across the record (Ying Zhang appears twice in the co-author list, possibly repeat collaboration or a deduplication artifact). On both 2026 arXiv papers the researcher sits in a middle author position without corresponding-author credit, consistent with contributing-specialist roles inside a small, stable document-AI team rather than independent or leading-author work.
Research keywords
Highlights
Controlled comparison of text vs. pixel inputs for document QA
MSBD: batching page-boundary decisions into one LLM call
A consistent cost-efficiency theme, with a shared rigor gap
Both reviewed papers attack inference cost in document AI — one on input representation, one on boundary batching — yet both stop short of significance testing on single-run results, a recurring methodological gap Evolve flags and an obvious next step for this line of work.
Timeline
- 2026Publishes an arXiv evaluation of text, image, and combined input representations for multimodal document QA under a 50-page cap, reporting accuracy gains for page images at growing latency cost.
- 2026Proposes MSBD for LLM-based page stream segmentation on arXiv, batching page-boundary decisions into a single call with accuracy held at a model-specific knee.
- 2026Both papers enter the Evolve catalog with medium reproducibility readiness; verdicts note controlled setups but single-run evidence without significance testing, alongside 5 recorded distinct coauthors.
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.
- Radar scores and metrics are computed from public bibliometric data only; the output, impact, and activity axes are unscored for lack of data.
- The profile is name-resolved from public bibliographic records; no ORCID, institution, or country is available, so identity confidence is limited.
- Field distribution, reproducibility mean, and timeline rest on only two reviewed 2026 arXiv preprints — a small, early sample that may not represent the full publication record.
- Reproducibility ratings and verdicts reflect Evolve's review of the two catalogued papers only; no peer-review outcomes or citation data are included.