NLP researcher at IIT Madras: retrieval-augmented generation, knowledge injection, structure-aware document retrieval
5
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
8
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—
No output score is available yet. This axis would measure publication volume and productivity over time; raw indicators exist (h-index 8, i10-index 6), but the underlying venue-level record is too incomplete for this profile to score.
Impact—
No impact score is available yet. This axis would capture citation-based influence, such as citation counts and field-normalized attention; beyond a small raw citation footprint, the data needed to compute it is currently insufficient.
Focus0/100
The focus dimension scores 0, and the raw data supports a genuinely broad footprint: output spans five OpenAlex topic clusters — topic modeling (48), NLP techniques (36), IR and search behavior (25), multimodal ML (17), semantic web (11) — while the field distribution splits exactly 50/50 between LLM adaptation/knowledge injection and information retrieval, with no single dominant line.
Collaboration56/100
Collaboration scores 56, a mid-range value. The record lists 16 distinct co-authors against 28 works in the last three years, and a stable recurring cluster — Sachindra Joshi, Vishwajeet Kumar, Meghanadh Pulivarthi, Vineet Kumar, Yatin Nandwani, Dinesh Raghu — appears in the frequent-collaborator list. All five catalog papers carry middle authorship, consistent with team-based rather than solo or PI-led work.
Activity98/100
Activity is the strongest axis at 98. The record lists 28 works in the last three years, and annual counts rise from 5 (2024) to 9 (2025) and 14 so far in 2026 — the highest yearly figure in a record reaching back to 2009, when only 1 work was recorded.
Reproducibility55/100
Reproducibility scores 55, drawn from Evolve's assessment of 2 reviewed papers (mean 55), both rated medium. The verdicts flag real caveats: STAIR's reported gains rest on synthetic queries and constrained decoding, and DKL's margins over retrieval-augmented baselines are often within noise — numbers appear plausible, but replication would need the released artifacts checked directly.
Research summary
Jaydeep Sen is a name-resolved research profile (OpenAlex; ORCID 0009-0009-9202-0300) affiliated with the Indian Institute of Technology Madras. The record's field distribution splits exactly 50/50 between NLP for LLM adaptation and knowledge injection, and NLP for information retrieval. OpenAlex topic weights agree: topic modeling (48), NLP techniques (36), IR and search behavior (25), multimodal machine learning (17), semantic web (11). Basic indicators are an h-index of 8 and an i10-index of 6; annual work counts climb from 1 in 2009 to 5 (2024), 9 (2025), and 14 in 2026, with 28 works in the last three years — the basis of an activity score of 98. Five 2026 arXiv papers sit in the Evolve catalog, all with Sen in a middle author position: STAIR (a table-of-contents-based generative retriever reported at 82.6 vs 76.9 R@1 over DSI), DKL (merging base-model knowledge vectors into instruction-tuned LLMs), plus KItCAT, SearchWiki, and Spatial Matryoshka Training. Evolve has reviewed two papers, both rated medium reproducibility (mean 55). Output and impact dimension scores are not yet computed; collaboration scores 56 across 16 distinct co-authors, and focus is 0 given the deliberately broad footprint.
Collaboration pattern
The record lists 16 distinct co-authors, twelve of whom recur as frequent collaborators, headed by Sachindra Joshi and including Vishwajeet Kumar, Meghanadh Pulivarthi, Vineet Kumar, Yatin Nandwani, and Dinesh Raghu. All five catalog papers place Sen in a middle author position with none corresponding, consistent with a steady contributor inside a stable NLP group; the collaboration dimension scores 56.
Research keywords
Highlights
STAIR: table of contents as the retrieval unit for generative retrieval
DKL: decoupled knowledge learning for instruction-tuned LLMs
A dense 2026 arXiv run across knowledge injection and document retrieval
Three further 2026 preprints — KItCAT (knowledge injection via input corruption), SearchWiki (building and navigating knowledge wikis for active information seeking), and Spatial Matryoshka Training (multi-granularity visual document retrieval) — extend the same two themes; none has an Evolve verdict yet.
Medium reproducibility across both reviewed papers
Evolve has reviewed 2 of Sen's papers with a mean reproducibility score of 55, both rated medium readiness — consistent with promising reported results whose replication depends on checking released artifacts and baseline margins.
Timeline
- 2009Earliest recorded publication year in the OpenAlex record, with 1 work.
- 20208 works recorded — the most productive year of the earlier phase of the record.
- 20259 works recorded, continuing a ramp-up that spans 28 works over the last three years.
- 202614 works so far, the record's highest annual count; five arXiv preprints enter the Evolve catalog (STAIR, DKL, KItCAT, SearchWiki, Spatial Matryoshka Training).
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.
- All bibliometrics come from a single OpenAlex snapshot; citation data other than h-index (8) and i10-index (6) is not available, so output and impact dimensions are unscored.
- Identity is name-resolved from public bibliographic data (OpenAlex plus ORCID 0009-0009-9202-0300); 'Jaydeep Sen' is not a unique name and the match relies on that record.
- The recorded country is listed as 'US' while the institution is IIT Madras; the discrepancy is retained exactly as provided and not reconciled.
- Three of the five catalog papers (KItCAT, SearchWiki, Spatial Matryoshka Training) have no Evolve verdict, keywords, or reproducibility rating yet, and no peer-review venue information is available for any catalog paper.