Microsoft (India) researcher bridging NLP and entity resolution through benchmark-first, reproducibility-oriented work
2
papers here
2
reviewed by Evolve
1
first-author
1
corresponding
—
works · OpenAlex
—
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—
Output (publication volume and productivity over time) is not yet scored: the underlying bibliometric coverage for this profile is insufficient. With only two Evolve-reviewed 2026 arXiv papers visible, no total publication count can be reported.
Impact—
Impact (citation uptake, downstream influence, and reuse of the benchmarks and methods) is not scored here. The catalogued papers are recent arXiv preprints and no citation counts are available in the input data.
Focus0/100
The computed focus score is 0, and the raw field distribution is consistent with that reading: the two catalogued papers split exactly 50% natural language processing and 50% data integration / entity resolution. Rather than concentrating on one field, Gupta's visible output so far sits evenly across two distinct research communities — peer-review generation with Llama-3.1-8B on one side, corporate-family entity resolution benchmarks on the other — making this a deliberately dual-track, early profile.
Collaboration39/100
Collaboration scores 39, based on a compact footprint of 6 distinct coauthors across the catalogued papers. The frequent-coauthor list — Maitreya Prafulla Chitale, Ketaki Mangesh Shetye, Yash More, Manav Chaudhary, Manish Shrivastava, and Vasudeva Varma — suggests a small, stable team. Gupta alternates roles: first and corresponding author on the entity-resolution benchmark, middle author on the peer-review generation study.
Activity—
Activity (publication cadence and year-over-year research tempo) is not scored. Both catalogued papers carry the same year, 2026, so there is no multi-year span from which to measure an activity trend.
Reproducibility78/100
Reproducibility is the strongest measured dimension, scoring 78 as a mean across 2 papers. The gradient is visible in per-paper readiness: the first-author corporate-family resolution benchmark rates high, consistent with Evolve's description of a rigorous, unusually honest public benchmark, while the multi-reference training study for peer-review generation rates medium, reflecting a controlled 8B setup whose SEA-S ground truth may bias results.
Research summary
Harshit Gupta is a researcher affiliated with Microsoft (India) whose visible work to date spans two complementary areas: natural language processing and data integration / entity resolution, which split his catalogued output evenly (50% each across the two papers Evolve has reviewed). The profile is built from a small but concrete base: two 2026 arXiv papers. As first and corresponding author, Gupta led 'Corporate-Family Resolution Is Not a String-Matching Problem', a public benchmark for corporate-family entity resolution stratified by name visibility; Evolve's review calls the benchmark rigorous and unusually honest, crediting the claim that corporate-family linkage fails at the blocking stage rather than at matching, while noting weak baselines and noisy ground truth. The second paper, a middle-author contribution, evaluates consolidation in multi-reference training for peer-review generation using Llama-3.1-8B and the MERC-36K dataset; Evolve notes the controlled 8B setup but flags possible bias in the SEA-S ground truth. Reproducibility is the strongest measured signal, with a mean readiness score of 78 across 2 papers. The collaboration footprint is compact — six distinct coauthors, including Manish Shrivastava and Vasudeva Varma — and output, impact, and activity metrics are not yet available, making this an early-stage profile.
Collaboration pattern
Gupta's collaboration footprint is compact: 6 distinct coauthors across the two catalogued papers, with a recurring group including Maitreya Prafulla Chitale, Ketaki Mangesh Shetye, Yash More, Manav Chaudhary, Manish Shrivastava, and Vasudeva Varma. The pattern suggests a small, stable team rather than wide external networking, consistent with the collaboration score of 39. Roles vary by project — he leads the entity-resolution benchmark as first and corresponding author while contributing as a middle author to the peer-review generation study.
Research keywords
Highlights
Public benchmark argues corporate-family resolution fails at blocking, not matching
Controlled 8B study tests consolidation in multi-reference training for peer reviews
Timeline
- 2026First-author, corresponding-author arXiv paper proposing a public corporate-family entity-resolution benchmark stratified by name visibility.
- 2026Co-authored (middle position) an arXiv study evaluating consolidation in multi-reference training for peer-review generation with Llama-3.1-8B and MERC-36K.
- 2026Both catalogued papers fall within 2026, establishing an early, dual-field profile spanning NLP and data integration / entity resolution.
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 are computed by Evolve from public bibliographic data; output, impact, and activity lack sufficient underlying data and are therefore unscored.
- Coverage is limited to the two arXiv papers reviewed by Evolve; publication counts, citation counts, and h-index are not available and are not estimated.
- No ORCID was provided; 'Harshit Gupta' is a common name, and this profile is name-resolved from public bibliographic data.
- Both catalogued papers are 2026 arXiv preprints; peer-reviewed venues and country-level affiliation detail are not captured in the data.