University of Surrey NLP researcher: LLM-augmented product search and interpretability of vision-language models.
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
1
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-09) 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 yet: the underlying publication-volume data is insufficient at this career stage. This axis would measure annual paper counts, venue distribution, and first-author share once a longer publication record accumulates.
Impact—
No impact score yet: citations have not accumulated on a record starting in 2024 (h-index 1, i10-index 0). This axis would measure citation counts, h-index trajectory, and uptake of the work by other researchers.
Focus0/100
The focus score is 0, reflecting an OpenAlex field distribution split exactly 50/50 between information retrieval (e-commerce product search) and AI interpretability / multimodal machine learning. Topic counts nonetheless suggest a coherent core — hate speech and cyberbullying detection (7), sentiment analysis and opinion mining (5), multimodal ML applications (5), topic modelling and NLP techniques (3 each) — so the diffuseness lies between fields rather than within them.
Collaboration44/100
Collaboration scores 44, based on 8 distinct co-authors to date. Both Evolve-reviewed papers are first-authored (non-corresponding), and the recurring partner list mixes Surrey NLP colleagues — Constantin Orasan, Diptesh Kanojia, Helen Treharne — with vision researcher Xiatian Zhu and peer co-authors Swapnil Bhosale, Samarth Agrawal, and Hadeel Sadany. The pattern is small-team, junior-first-author work inside a departmental cluster rather than large consortium science.
Activity75/100
Activity scores 75, the strongest signal in this profile: 12 works over the last three years, with OpenAlex yearly counts climbing from 2 (2024) to 3 (2025) to 7 (2026) — a near-tripling in the most recent year. Both 2026 catalogue entries are arXiv preprints, indicating the current pace is driven by rapid, first-authored preprint output.
Reproducibility55/100
Reproducibility scores 55 (reproMean over 2 reviewed papers). Evolve rates both 2026 preprints at medium readiness: the product-search paper reports concrete nDCG@10 numbers (0.9382 oracle vs 0.9258) but on only 75 test queries without significance testing; the interpretability paper's causal claims rest on 5–15-example activation-patching interventions. Candid limitation statements in both help, but small evaluation sets temper confidence.
Research summary
Girish A. Koushik is an early-career researcher at the University of Surrey (GB) whose OpenAlex-indexed work splits almost evenly between two fields: information retrieval for e-commerce product search and AI interpretability within multimodal machine learning. His visible record is young but accelerating — 12 works in the last three years, with yearly counts rising from 2 (2024) to 3 (2025) to 7 (2026) — produced with 8 distinct co-authors, and an h-index of 1 (i10-index 0). Two first-authored 2026 arXiv preprints anchor his Evolve-reviewed catalog. 'Beyond Relevance...' introduces LLM-augmented, human-reannotated structured semantic supervision for product search, reporting oracle nDCG@10 of 0.9382 versus 0.9258 on the ESCI benchmark while candidly noting only 75 test queries and no significance testing. 'Decodable but Misrouted...' shows sparse-autoencoder probes decoding harmful-meme evidence that Gemma-3 and Qwen3.5 fail to route to outputs — unusually candid and carefully scoped, though causal evidence rests on 5–15-example interventions. Evolve rates both papers medium on reproducibility readiness (reproMean 55 across 2 papers). His topic profile — hate speech and cyberbullying detection (7), sentiment analysis (5), multimodal ML applications (5), topic modelling and NLP techniques (3 each) — points to a consistent interest in content understanding and model behaviour.
Collaboration pattern
Koushik's collaboration footprint is compact: 8 distinct co-authors, with a collaboration dimension score of 44. His recurring partners cluster at Surrey — NLP researchers Constantin Orasan, Diptesh Kanojia, and Helen Treharne, plus vision researcher Xiatian Zhu — alongside peer co-authors Swapnil Bhosale, Samarth Agrawal, and Hadeel Sadany. Both Evolve-reviewed papers are first-authored but not corresponding, consistent with a junior researcher doing leading work within an established departmental group rather than leading his own team yet.
Research keywords
Highlights
Structured semantic supervision lifts product search re-ranking
Sparse features expose a readout gap in vision-language models
Timeline
- 2024Earliest OpenAlex-recorded output: 2 works appear in 2024, the starting point of the visible publication record.
- 2025Output grows modestly to 3 works, building the base of the NLP and multimodal topic profile recorded by OpenAlex.
- 2026Output accelerates to 7 works in a single year; the three-year total reaches 12 works, the strongest activity signal in the profile.
- 2026Two first-authored arXiv preprints enter the Evolve catalog — one on LLM-augmented product search, one on sparse-feature readout gaps in vision-language models — both rated medium on reproducibility readiness.
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 are derived from OpenAlex only; identity is name-resolved from public bibliographic data.
- Citation-based output and impact could not be scored: the record starts in 2024, with h-index 1 and i10-index 0.
- Only the two 2026 arXiv preprints have Evolve reviews; topic and field distributions cover the broader OpenAlex record, which includes works not in this catalog.
- Both catalogued papers are preprints; peer-review outcomes may change the picture.