Nikhil Singh

Dartmouth College

Unclaimed · AI-structured profileORCID OpenAlex

NLP and neurosymbolic AI researcher at Dartmouth working on LLM information extraction and Bayesian-network benchmarks

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-17) 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—

Output (publication volume and productivity over time) is not scored: the current snapshot lists only three 2026 arXiv preprints, with no verified publication counts or venue history, so there is insufficient data to compute this axis.

Impact—

Impact (citation counts, h-index, and influence on later work) cannot be scored: no citation data was provided for these recent arXiv preprints, so this axis is reported as missing rather than estimated.

Focus100/100

The focus score is 100, the maximum, because every classified paper in the catalog falls into a single field — natural language processing / neurosymbolic AI (field distribution: 1 field). The three preprints are thematically contiguous: information extraction, text-to-Bayesian-network extraction, and world models in LLM agents all sit within LLM-centered NLP, so the current record shows no field dispersion.

Collaboration46/100

The collaboration score is 46, derived from 9 distinct coauthors in the snapshot. Recurring collaborators include Soroush Vosoughi, Omar Sharif, Ruize Xu, Xiao Yu, Yujin Tang, Chenming Shang, Amartya Bhattacharya, and Neeti Pokhriyal. On the reviewed papers he takes middle and last author positions, never corresponding author, indicating team-embedded contributions rather than solo or lead-corresponding output.

Activity—

Activity (recent publication cadence and year-over-year research rhythm) is not scored: the snapshot covers three preprints all dated 2026, with no multi-year record to measure sustained output.

Reproducibility55/100

The reproducibility score is 55, computed from a single assessed paper (reproMean 55, reproPapers 1). Both rated papers are marked medium readiness: PRISM-BN ships a controlled corpus and benchmark whose reference networks are LLM-generated, and the learned-queries paper reports convincing oracle-conditioned gains (+18.6 F1) that are not statistically tested. The evidence base is narrow, so this score should be read as provisional.

Research summary

Nikhil Singh is a researcher at Dartmouth College working at the intersection of natural language processing and neurosymbolic AI. The Evolve catalog associates him with three 2026 arXiv preprints. The first, PRISM-BN, contributes a controlled corpus and benchmark for text-to-parameterized Bayesian network extraction; Evolve's review describes it as carefully built and honest, while flagging that the reference Bayesian networks are LLM-generated and not externally validated. The second, Improving Information Extraction with Learned Queries, argues that question design can beat model scaling in LLM-based information extraction; Evolve finds the oracle-conditioned gain of +18.6 F1 convincing, but notes that deployable gains are modest and untested statistically. The third studies how world models and policies compose in LLM agents through a joint spectral and behavioral account, and has not yet been reviewed by Evolve. Radar metrics reflect an early, tightly consolidated profile: a focus score of 100, with all classified catalog output in a single field (natural language processing / neurosymbolic AI); a collaboration score of 46 based on 9 distinct coauthors; and a reproducibility score of 55 derived from one assessed paper. He appears in middle and last author positions on the reviewed papers, none as corresponding author.

Collaboration pattern

Snapshot data records 9 distinct coauthors. Recurring collaborators include Soroush Vosoughi, Omar Sharif, Ruize Xu, Xiao Yu, Yujin Tang, Chenming Shang, Amartya Bhattacharya, and Neeti Pokhriyal. On the two reviewed papers he holds middle and last author positions, neither as corresponding author, consistent with contributing inside a team rather than leading as corresponding author. The collaboration score of 46 sits mid-range, reflecting a moderate-size, recurring coauthor circle associated with Dartmouth.

Research keywords

Information extractionQuestion generationPrompt optimizationBayesian networksText-to-BN extractionProbabilistic graphical modelsCPD elicitationNeurosymbolic AIEvent argument extractionClinical NLPBenchmark datasetsLLM agents

Highlights

PRISM-BN: a controlled benchmark for text-to-parameterized Bayesian network extraction
Evolve's review calls PRISM-BN a carefully built, honest benchmark and corpus for text-to-parameterized-BN extraction, with the key caveat that the reference Bayesian networks are LLM-generated and not externally validated. Reproducibility readiness is rated medium.
Learned queries: question design as a lever in LLM information extraction
This paper argues that question design can beat model scaling in LLM information extraction; Evolve finds the oracle-conditioned +18.6 F1 gain convincing, while noting that deployable gains are modest and statistically untested. Reproducibility readiness is rated medium.
How world models and policies compose in LLM agents
A third 2026 preprint offers a joint spectral and behavioral account of how world models and policies compose in LLM agents. Evolve has not yet reviewed it, so no verdict, keywords, or reproducibility rating exists for this work.

Timeline

2026Released PRISM-BN, a controlled corpus and benchmark for text-to-parameterized Bayesian network extraction (arXiv, middle authorship).
2026Posted Improving Information Extraction with Learned Queries, arguing question design can beat model scaling in LLM information extraction (arXiv, last authorship).
2026Released a joint spectral and behavioral study of how world models and policies compose in LLM agents (arXiv, last authorship; not yet reviewed by Evolve).

Papers in the Evolve catalog

Frequent co-authors

Soroush Vosoughi×2 Omar Sharif×1 Ruize Xu×1 Xiao Yu×1 Yujin Tang×1 Chenming Shang×1 Amartya Bhattacharya×1 Nikhil Singh×1 Neeti Pokhriyal×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.

Bibliometric radar scores are computed by Evolve from OpenAlex-derived bibliometric data; coverage outside this snapshot is not reflected.
Output, impact, and activity dimensions are reported as missing because no publication counts, citation data, or multi-year records were available.
All three catalog papers are 2026 arXiv preprints; peer-review status, venues, and citation counts are unknown, and only one paper carries a reproducibility score.
"Nikhil Singh" is a common name; this profile is name-resolved from public bibliographic data and anchored to ORCID 0000-0003-4465-6469.