Vishwajeet Kumar
👨‍🔬 Vishwajeet Kumar
🏛️
Position Staff Research Scientist
📚
Citations 4,600+ on Scholar
🔬 IBM Research AI · Bengaluru, India

Vishwajeet
Kumar

Staff Research Scientist — Speech & Language Group
IBM Research Lab, India

Researcher focused on Large Language Models for RAG and text embedding for information retrieval. Joint Ph.D. from IIT Bombay & Monash University. Co-author of IBM's open-source Granite Embedding models.

30+
Papers
10+
Patents
4.6K+
Citations
📖 Google Scholar IBM Research ↗ LinkedIn ↗
Scroll
Research Focus

What I Work On

My research sits at the intersection of language models, structured reasoning, and efficient information retrieval for enterprise AI.

🔍 Retrieval Augmented Generation ❓ Question Answering & Generation 🌐 Multilingual NLP · Indic Languages 🏗️ Knowledge Graphs & Ontologies 🧠 Large Language Models 📊 Table Understanding 🔗 Code-Mixed NLP (Hinglish) ⚡ Embedding Models & Dense Retrieval
🤖
Latest Highlight
IBM Granite Embedding R2 — Aug 2025
Co-authored the Granite Embedding model family — open-source encoder models for dense/sparse retrieval released under Apache 2.0, outperforming similar-sized public models on IBM retrieval benchmarks.
R1 Paper → R2 Paper →
Live Feed

Research Updates 10

Papers, patents, models, and milestones — as they happen.

📄
New Paper
LMK > CLS: Landmark Pooling for Dense Embeddings
Introduces Landmark (LMK) pooling for sequence encoders — partitions input into chunks with interleaved landmark tokens, improving long-context retrieval substantially while matching existing methods on short-context tasks. A scalable alternative to [CLS] and mean pooling.
arXiv:2601.21525 · Jan 2026 Read Paper →
JAN 2026
📄
New Paper
Influence Guided Sampling for Domain Adaptation of Text Retrievers
Proposes Inf-DDS — a novel RL-driven training data sampling framework for embedding models that adaptively reweights datasets using influence-based rewards. Achieves 5.03 NDCG@10 improvement on multilingual bge-m3 at 1.5–4× lower GPU cost than gradient-based baselines.
arXiv:2601.21759 · Jan 2026 Read Paper →
JAN 2026
🔒
Patent Granted
US Patent 12639351 Granted — Table-of-Contents in Differential Search Index
New US patent granted for a system and methods that induce Table-of-Contents knowledge into a Differential Search Index — advancing structured document understanding and retrieval for enterprise AI.
US 12639351 · May 26, 2026 View Patent →
MAY 2026
🤖
Model Released
Granite Embedding R2 Models
Co-authored the next-generation IBM Granite R2 embedding models — ModernBERT-based architecture trained on 2T tokens, with extended context length and improved inference optimizations. Released under Apache 2.0 for commercial & research use.
arXiv:2508.21085 Read Paper →
AUG 2025
📄
Technical Report
Granite Embedding Models — IBM's Open Retrieval Family
Released the Granite Embedding family — encoder models for dense and sparse retrieval with English and multilingual capabilities. Uses retrieval-oriented pretraining, contrastive finetuning, knowledge distillation, and model merging.
arXiv:2502.20204 · Feb 2025 Read Paper →
FEB 2025
📄
Conference Paper
MILU: A Multi-task Indic Language Understanding Benchmark
Accepted at NAACL 2025. A comprehensive multi-task benchmark for evaluating understanding across major Indic languages, advancing multilingual NLP evaluation beyond English.
NAACL 2025
2025
📄
Short Paper
Benchmarking Zero-Shot Hindi Retrieval with Hindi-BEIR & NLLB-E5
Accepted at NAACL 2025. Introduces Hindi-BEIR for Hindi information retrieval benchmarking alongside a zero-shot retrieval model for low-resource Hindi search.
NAACL 2025
2025
🔒
Patent Granted
US Patent 12411875 Granted
Latest US patent granted, covering NLP-based question answering systems — part of a growing IP portfolio of 10+ patents and disclosures in AI language systems.
US 12411875 · Sep 8, 2025
SEP 2025
📄
Demo Paper
PrimeQA: State-of-the-Art Multilingual Question Answering
Accepted at ACL 2023. IBM's open-source unified framework for multilingual QA research — enabling researchers to build, train, and evaluate across languages at scale.
ACL 2023 Read Paper →
2023
📄
Conference Paper
Hinglish Code-Mixed NLP via Meta-Learning & Embedding Resources
Accepted at EMNLP 2022. A meta-learning framework leveraging Hindi and English language resources significantly improves performance on Hinglish NLP tasks using the GLUECoS benchmark.
EMNLP 2022
2022
Research Output

Selected Publications

30+ peer-reviewed papers across ACL, EMNLP, NAACL, SIGIR, PAKDD, ISWC, IJCAI and more.

2027
🚀
Upcoming in 2027
Papers and research outputs in progress. Check back here for updates — or follow on Google Scholar for the latest.
🔍 RAG Research ⚡ Embedding Models 🌐 Multilingual NLP
2026
arXiv 2026
LMK > CLS: Landmark Pooling for Dense Embeddings
Meet Doshi, Aashka Trivedi, Vishwajeet Kumar, Parul Awasthy, Yulong Li, Jaydeep Sen, Radu Florian, Sachindra Joshi
Introduces Landmark (LMK) pooling — a new sequence encoding strategy that partitions input into chunks with landmark tokens, improving long-context retrieval while preserving local salient signals.
arXiv →
arXiv 2026
Influence Guided Sampling for Domain Adaptation of Text Retrievers
Meet Doshi, Vishwajeet Kumar, Yulong Li, Jaydeep Sen
Proposes Inf-DDS, a reinforcement learning framework that adaptively reweights training datasets using influence-based reward signals — achieving up to 5.03 NDCG@10 improvement with 1.5–4× lower GPU cost.
arXiv →
2025
NAACL 2025
MILU: A Multi-task Indic Language Understanding Benchmark
Sshubam Verma, Mohammed Safi Ur Rahman, et al. (incl. V. Kumar)
Paper →
NAACL 2025
Benchmarking Zero-Shot Hindi Retrieval with Hindi-BEIR and NLLB-E5
Arkadeep Acharya, Rudra Murthy, et al. (incl. V. Kumar)
Paper →
arXiv 2025
Granite Embedding Models (R1)
Parul Awasthy, Aashka Trivedi, ..., Vishwajeet Kumar, ..., Radu Florian
PDF →
arXiv 2025
Granite Embedding R2 Models
Parul Awasthy, Aashka Trivedi, ..., Vishwajeet Kumar, ..., Radu Florian
PDF →
2023
ACL 2023
PrimeQA: The Prime Repository for State-of-the-Art Multilingual Question Answering
Avi Sil, Jaydeep Sen, et al. (incl. V. Kumar)
Paper →
2022
EMNLP 2022
On Utilizing Matrix and Embedding Language Resources to Improve Downstream Tasks in Hinglish
Vishwajeet Kumar, Rudra Murthy Venkataramana, Tejas Indulal Dhamecha
Paper →
2021
EMNLP 2021
Topic Transferable Table Question Answering
Saneem Ahmed Chemmengath, Vishwajeet Kumar, Samarth Bharadwaj, et al.
arXiv →
SIGIR 2021
Select, Substitute, Search: Knowledge-Augmented Visual QA Benchmark
Aman Jain, Mayank Kothyari, V. Kumar, Preethi Jyothi, et al.
PDF →
NAACL 2021
Capturing Row and Column Semantics in Transformer QA over Tables
Michael Glass, Mustafa Canim, Alfio Gliozzo, Saneem Chemmengath, V. Kumar, et al.
PDF →
EACL 2021
Meta-Learning for Effective Multi-task and Multilingual Modelling
Ishan Tarunesh, Sushil Khyalia, V. Kumar, Ganesh Ramakrishnan, Preethi Jyothi
PDF →
2019
ACL 2019
Cross-Lingual Training for Automatic Question Generation
Vishwajeet Kumar, Nitish Joshi, Arijit Mukherjee, Ganesh Ramakrishnan, Preethi Jyothi
PDF →
IJCAI 2019
Neural Program Induction for KBQA Without Gold Programs
Amrita Saha, Ghulam Ansari, V. Kumar, Karthik Shankaranarayanan, et al.
PDF →
EMNLP 2019
ParaQG: A System for Generating Questions and Answers from Paragraphs
Vishwajeet Kumar et al.
PDF →
ISWC 2019
Difficulty-Controllable Multi-hop Question Generation from Knowledge Graphs
Vishwajeet Kumar et al.
Earlier
NAACL 2018
Entity Resolution & Location Disambiguation in Ancient Hindu Temples Domain
Ayush Maheshwari, V. Kumar, Ganesh Ramakrishnan, Sakethanath Jagarlapudi
PDF →
PAKDD 2018
Automating Reading Comprehension by Generating Question and Answer Pairs
Vishwajeet Kumar, Kreeti Boorla, Ganesh Ramakrishnan, Yuan Fang Li
PDF →
EMNLP 2016
Towards Semi-Automatic Generation of Proposition Banks for Low-Resource Languages
Alan Aikbik, V. Kumar, Yunyao Li
VLDB 2016
Civique: Using Social Media to Detect Urban Emergencies
Diptesh Kanojia, V. Kumar, Krithi Ramamritham
IP Portfolio

Patents & Disclosures

11+ patents and disclosures in NLP, structured data reasoning, and AI language systems.

US 12639351 · May 26, 2026
System and Methods for Inducing Table-of-Contents Knowledge in Differential Search Index
Granted View →
US 12411875 · Sep 8, 2025
Question Generation & Answering System (NLP)
Granted 🇺🇸 United States
US 12210538 · Jan 27, 2025
Multi-instance, Multi-answer Training For Table And Text Question Answering
Granted View →
US 12182508 · Dec 30, 2024
Natural Language Question Answering Using Non-relational Tables
Granted View →
US 12050877 · Jul 29, 2024
Contextual Dialogue Framework Over Dynamic Tables
Granted View →
IBM Disclosure · NLP Systems
Question Generation System for Structured Tables and Passages
Disclosed 🔒 IBM
Full Portfolio
11+ Patents & Disclosures
Spanning NLP, table reasoning, question answering, knowledge graphs, and enterprise AI systems.
Presentations

Talks & Workshops

Invited talks, tutorials, and workshop contributions at leading AI/NLP venues worldwide.

2023
PrimeQA: Multilingual Question Answering at Scale
ACL 2023 Demo Session · Toronto, Canada
2022
Code-switched NLP and Meta-Learning for Hinglish
EMNLP 2022 · Abu Dhabi, UAE
2021
Table Question Answering with Topic Transfer
EMNLP 2021 · Punta Cana, Dominican Republic
2021
Transformer-based Reasoning over Structured Tables
NAACL 2021 · Online
2019
Cross-Lingual Training for Automatic Question Generation
ACL 2019 · Florence, Italy
2019
Multi-hop Question Generation from Knowledge Graphs
ISWC 2019 · Auckland, New Zealand
2017
Dictionary Generalization Across Languages
Amazon India AI Summit 2017 · Poster Presentation
Get in Touch

Contact

Open to research collaborations, academic discussions, and opportunities in NLP and AI.

📍
Location
IBM Research India · Bengaluru, Karnataka
✉️
Email
[firstname][lastname]86 [at] gmail [dot] com
(firstname + lastname + 86)
🎓
Education
Ph.D. — IIT Bombay & Monash University (2020)
Computer Science & Engineering
🏢
Affiliation
IBM Research India · Speech & Language Group
Research Interests
🔍 Retrieval Augmented Generation (RAG)
❓ Question Answering & Generation
🌐 Multilingual & Indic NLP
🧠 Large Language Models
⚡ Embedding Models & Dense Retrieval
🏗️ Knowledge Graphs & Structured Reasoning
🔑
Owner Access

Enter your secret key to post a new update.

Only the site owner has this key.