Industry researcher at Amazon formalizing verifiable human-LLM oversight and epistemic warrant for AI advice
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
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 scored. The snapshot lacks a reliable cumulative output measure for this identity (no ORCID-linked record; h-index and i10-index are 0), so total publication count and venue mix cannot be estimated.
Impact—
Not scored. This axis would measure citation uptake — citations, h-index, i10-index — but the OpenAlex record reports h-index 0 and i10-index 0, insufficient to characterize impact. No estimate is made.
Focus0/100
The focus score is 0, reflecting an even split of the field distribution — 50% information systems / AI evaluation and 50% natural language processing / human-AI interaction (verification theory) — rather than a single dominant field. Topic counts reinforce this breadth: explainable AI (4), topic modeling (3), adversarial robustness in ML (3), ethics and social impacts of AI (2), and NLP techniques (1). Work is thematic but deliberately multi-field.
Collaboration28/100
Collaboration scores 28 on a 0-100 scale, grounded in only 3 distinct coauthors in the bibliometric record. Recurring collaborators are Shai Vardi, Baotong Zhang, and Dean Foster. Both reviewed 2026 papers place Sedoc in the last author position without a corresponding-author flag, suggesting a senior contributor on small teams rather than large consortium work. The sample is small, so this picture may shift quickly with further publications.
Activity67/100
Activity scores 67, supported by counts-by-year data showing 3 works in 2025 and 6 in 2026 — a doubling of annual output — with 9 works inside the recent 3-year window. Both cataloged papers are 2026 arXiv preprints, indicating a current publishing cadence centered on LLM evaluation and oversight. All recorded output falls within 2025-2026, so this score reflects a short, recent burst of activity rather than an extended track record.
Reproducibility78/100
Reproducibility scores 78 (the mean across 2 assessed papers). Evolve rates the reliance-certificate paper high and the interactive-proof paper medium: the latter pairs a rigorous soundness proof with task-specific bounds that leave practical applicability open, while the former shows careful multi-model validation. The 78 mean signals consistently documented, validation-oriented methods, though the evidence base of two reviewed papers is narrow.
Research summary
João Sedoc is an industry-based researcher at Amazon (United States) whose recent work spans information systems / AI evaluation and natural language processing / human-AI interaction with a verification-theory orientation. The OpenAlex-resolved profile (no linked ORCID) records 9 works in the recent 3-year window, split as 3 publications in 2025 and 6 in 2026, with h-index and i10-index listed as 0. Topic frequency concentrates on explainable artificial intelligence (4), topic modeling (3), adversarial robustness in machine learning (3), ethics and social impacts of AI (2), and NLP techniques (1). Evolve has reviewed two 2026 arXiv papers, both with Sedoc in the last author position and neither flagged as corresponding author. One formalizes human-LLM review as interactive proof, with a rigorous soundness proof but task-specific bounds that leave practical applicability largely open; the other proposes a four-tier reliance certificate for LLM recommendations when ground truth is unavailable, rated high for reproducibility. The overall picture is an early-stage, verification-focused agenda: output is recent and concentrated in 2025-2026, the coauthor network is narrow at 3 distinct collaborators, and citation-based impact cannot yet be scored.
Collaboration pattern
The bibliometric record lists only 3 distinct coauthors, yielding a collaboration score of 28/100. Recurring collaborators are Shai Vardi, Baotong Zhang, and Dean Foster. Both reviewed 2026 papers place Sedoc in the last author position without corresponding-author status, consistent with a senior contributor on small, focused teams rather than large multi-institution consortia. With so small a sample, the collaboration pattern should be read as provisional and likely to evolve as further work appears.
Research keywords
Highlights
Human-LLM review formalized as interactive proof
Four-tier reliance certificate for LLM advice without ground truth
Timeline
- 2025The bibliometric record shows 3 publications in 2025 (counts-by-year), the earliest activity captured in the current snapshot.
- 2026Annual output doubles to 6 publications in 2026 per counts-by-year, bringing the recent 3-year total to 9 works.
- 2026arXiv preprint 'Human-LLM Deliberation as Interactive Proof' enters the Evolve catalog; reproducibility is rated medium.
- 2026arXiv preprint 'Epistemic Warrant for LLM Recommendations' enters the Evolve catalog; reproducibility is rated high.
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.
- Bibliometrics come from an OpenAlex-resolved identity only; no ORCID is linked, so name resolution relies on public bibliographic matching.
- Citation impact is unscored: the snapshot lists h-index 0 and i10-index 0, which may not capture work indexed under other identity variants.
- Only two Evolve-reviewed papers underlie the qualitative assessments, so reviewed coverage is narrow relative to the 9 recent works recorded.
- Affiliation (Amazon, United States) and author positions are taken from the input record and not independently verified.