BharatGen

Research papers

Peer-reviewed work and preprints from BharatGen and our consortium on Indian-language AI, speech, vision and efficient training.

33papers since 2025
25peer-reviewed
15venues, including NeurIPS, ICLR, CVPR and ACL
5research areas

All papers

Search, filter by area or type, and copy a citation in one click.

202611 papersShowHide
Oct 2026
EMNLP 2026 (Findings)Benchmarks

BhashaBench V1: A Comprehensive Benchmark for the Quadrant of Indic Domains

Vijay Devane, Mohd Nauman, Bhargav Patel and 13 others

Summary and full author list

A bilingual benchmark of 74,166 question-answer pairs in English and Hindi, drawn from government and domain exams in agriculture, law, finance and Ayurveda.

Vijay Devane, Mohd Nauman, Bhargav Patel, Aniket Mahendra Wakchoure, Yogeshkumar Sant, Shyam Pawar, Viraj Thakur, Ananya Godse, Sunil Patra, Neha Maurya, Suraj Racha, Nitish Kamal Singh, Piyush Sawarkar, Kundeshwar Vijayrao Pundalik, Rohit Saluja, Ganesh Ramakrishnan

Paper ↗
Sep 2026
ICDAR 2026 (oral)Vision

REPLICA: An Agentic Framework for Visually Faithful Document Reconstruction

Raghuveer R, Anirudh Srinivasan, Venkata Kesav Venna and 4 others

Summary and full author list

An agentic framework that rebuilds a document page as web-ready HTML, keeping both its content and how it looks.

Raghuveer R, Anirudh Srinivasan, Venkata Kesav Venna, Shanmukha Sreevatsa Tallapragada, Aryan Jain, Sahithi Kukkala, Ravi Kiran Sarvadevabhatla

Paper ↗
Sep 2026
ECCV 2026Vision

DocLayout‑VL: A Foundational Model for Hierarchical, Open‑Set and Promptable Document Layout Segmentation

Venkata Kesav Venna, Srihari Bandarupalli, Anirudh Srinivasan and 3 others

Summary and full author list

Treats document layout as a promptable, open-vocabulary task: it maps a page into a hierarchy of regions and gives each region a pixel-level mask.

Venkata Kesav Venna, Srihari Bandarupalli, Anirudh Srinivasan, Raghuveer R, Sai Madhusudan Gunda, Ravi Kiran Sarvadevabhatla

Paper ↗
Sep 2026
ECCV 2026 (spotlight)Vision

DoCoG: Mask‑based Multi‑Type Grounded Chain‑of‑Thought for Document QA

Sai Madhusudan Gunda, Jyothi Swaroopa Jinka, Hrithik Sagar Rachakonda and 4 others

Summary and full author list

A document question answering model that reasons step by step and marks the supporting evidence for each step, across text, tables and graphical elements such as chart legends and flowchart nodes. Presented as a spotlight at ECCV 2026, and also accepted at three ECCV workshops: eXCV, MUCG and UniWorld.

Sai Madhusudan Gunda, Jyothi Swaroopa Jinka, Hrithik Sagar Rachakonda, Aryan Jain, Venkata Kesav Venna, Anirudh Srinivasan, Ravi Kiran Sarvadevabhatla

Paper ↗
Jun 2026
CVPR 2026Vision

M3Grounder: Mask‑Based Multi‑Span and Multi‑Granular Grounding for Document QA

Venkata Kesav Venna, Sai Madhusudan Gunda, Jyothi Swaroopa Jinka and 3 others

Summary and full author list

Answers questions about a document and links each part of the answer to its evidence on the page with pixel-level masks, at phrase, line and block level.

Venkata Kesav Venna, Sai Madhusudan Gunda, Jyothi Swaroopa Jinka, Hrithik Sagar Rachakonda, Anirudh Srinivasan, Ravi Kiran Sarvadevabhatla

Paper ↗
Apr 2026
AISTATS 2026Training methods

UniPROT: Uniform Prototype Selection via Partial Optimal Transport with Submodular Guarantees

Prateek Chanda, Prayas Agrawal, Karthik S. Gurumoorthy and 3 others

Summary and full author list

Addresses imbalanced domain mixtures for language model training via Partial Optimal Transport with uniform weightage across domains to mitigate class imbalance.

Prateek Chanda, Prayas Agrawal, Karthik S. Gurumoorthy, Ganesh Ramakrishnan, Bamdev Mishra, Pratik Jawanpuria

Paper ↗
Mar 2026
LoResLM @ EACL 2026Text

Do Tokenizers Fail on Informal Hindi Expressions? Evidence from Static, Downstream, and Robustness Analyses

Manikandan Ravikiran, Tanmay Tiwari, Vibhu Gupta and 3 others

Summary and full author list

Studies how existing multilingual tokenizers handle informal Hindi expressions via static, downstream, and robustness analyses, revealing significant gaps on romanised text.

Manikandan Ravikiran, Tanmay Tiwari, Vibhu Gupta, Rakesh Prakash, Rohit Saluja, Shayan Mohanty

Paper ↗
Mar 2026
EACL 2026Speech

Post‑ASR Correction in Hindi: Comparing Language Models and Large Language Models in Low‑Resource Scenarios

Rishabh Kumar, Amrith Krishna, Ganesh Ramakrishnan and 1 other

Summary and full author list

Shows that smaller fine-tuned models like ByT5 and mT5 outperform larger LLMs for Hindi ASR post-correction, exhibiting inverse-scaling under zero-shot ICL and generalising across Marathi and Telugu.

Rishabh Kumar, Amrith Krishna, Ganesh Ramakrishnan, Preethi Jyothi

Paper ↗
Mar 2026
WACV 2026Vision

Tables Decoded: DELTA for Structure, TARQA for Understanding

Jahanvi Rajput, Dhruv Kudale, Saikiran Kasturi and 2 others

Summary and full author list

Converts table images to structured text for precise table recognition and visual question answering.

Jahanvi Rajput, Dhruv Kudale, Saikiran Kasturi, Utkarsh Verma, Ganesh Ramakrishnan

Paper ↗
Jan 2026
ICLR 2026Training methods

Sculpting Subspaces: Constrained Full Fine‑Tuning in LLMs for Continual Learning

Nikhil Shivakumar Nayak, Krishnateja Killamsetty, Ligong Han and 8 others

Summary and full author list

Proposes an adaptive SVD-based full fine-tuning method that constrains updates to orthogonal subspaces of prior tasks, achieving near-zero catastrophic forgetting without added parameters.

Nikhil Shivakumar Nayak, Krishnateja Killamsetty, Ligong Han, Abhishek Bhandwaldar, Prateek Chanda, Kai Xu, Hao Wang, Aldo Pareja, Oleg Silkin, Mustafa Eyceoz, Akash Srivastava

Paper ↗
202522 papersShowHide
Dec 2025
IJCNLP-AACL 2025 (Findings)Speech

HiLearners: Non‑Native Spoken Hindi Error Correction

Sourava Kumar Behera, Rohit Saluja

Summary and full author list

Addresses automatic error correction in spoken Hindi produced by non-native speakers, covering phonological, lexical, and grammatical error types across diverse language backgrounds.

Sourava Kumar Behera, Rohit Saluja

Paper ↗
Dec 2025
ICDAR 2025Vision

Towards Scene Text Recognition in Rainy Weather Conditions

Anandita Jamwal, Lalithya Koneti, Manikandan Ravikiran and 2 others

Summary and full author list

Investigates robustness of scene text recognition models under rainy conditions and proposes methods to handle rain-induced visual degradation in natural scene images.

Anandita Jamwal, Lalithya Koneti, Manikandan Ravikiran, Dinesh Singh, Rohit Saluja

Paper ↗
Dec 2025
BHASHA @ IJCNLP-AACL 2025Vision

Multi‑Feature Graph Convolution Network for Hindi OCR Verification

Shikhar Dubey, Krish Mittal, Sourava Kumar Behera and 4 others

Summary and full author list

Uses graph convolutional networks combining multiple visual features to verify and correct OCR outputs in Hindi, improving reliability on complex Devanagari script.

Shikhar Dubey, Krish Mittal, Sourava Kumar Behera, Manikandan Ravikiran, Nitin Kumar, Saurabh Shigwan, Rohit Saluja

Paper ↗
Nov 2025
EMNLP 2026 (accepted)Benchmarks

IndicParam: Benchmark to Evaluate LLMs on Low‑Resource Indic Languages

Ayush Maheshwari, Kaushal Sharma, Vivek Patel and 1 other

Summary and full author list

A 13K-question benchmark across 11 low-resource Indic languages showing current LLMs perform poorly (max ~58%), highlighting limits in cross-lingual understanding and reasoning.

Ayush Maheshwari, Kaushal Sharma, Vivek Patel, Aditya Maheshwari

Paper ↗
Sep 2025
NeurIPS 2025Training methods

Bandit Guided Submodular Curriculum for Adaptive Subset Selection

Prateek Chanda, Prayas Agrawal, Saral Sureka and 3 others

Summary and full author list

Recasts curriculum learning as a bandit over submodular selectors and proposes ONLINESUBMOD, a no-regret online greedy method that uses validation rewards to outperform prior approaches.

Prateek Chanda, Prayas Agrawal, Saral Sureka, Lokesh Reddy Polu, Atharv Kshirsagar, Ganesh Ramakrishnan

Paper ↗
Sep 2025
arXiv 2025Training methods

TaskMixPGM: Task Mixtures via Probabilistic Graphical Modelling for Language Model Finetuning

Prateek Chanda, Saral Sureka, Parth Pratim Chatterjee and 3 others

Summary and full author list

Introduces TASKPGM, an MRF-based framework that optimises task mixture proportions via behavioural divergences, yielding a closed-form, theoretically grounded solution that improves LLM finetuning performance and interpretability.

Prateek Chanda, Saral Sureka, Parth Pratim Chatterjee, Krishnateja Killamsetty, Nikhil Shivakumar Nayak, Ganesh Ramakrishnan

Paper ↗
Sep 2025
NeurIPS 2025Training methods

FairPO: Robust Preference Optimization for Fair Multi‑Label Learning

Soumen Kumar Mondal, Prateek Chanda, Akshit Varmora and 1 other

Summary and full author list

Introduces FairPO, a preference-driven, group-robust framework that boosts underperforming labels via DPO-style ranking corrections while preserving others through constrained optimisation and adaptive balancing.

Soumen Kumar Mondal, Prateek Chanda, Akshit Varmora, Ganesh Ramakrishnan

Paper ↗
Aug 2025
ICNLSP 2025Text

HiSlang‑4.9k: A Benchmark Dataset for Hindi Slang Detection and Identification

Tanmay Tiwari, Vibhu Gupta, Manikandan Ravikiran and 1 other

Summary and full author list

Introduces a 4,900-instance annotated dataset for detecting and identifying slang expressions in Hindi text, supporting informal language understanding in social media contexts.

Tanmay Tiwari, Vibhu Gupta, Manikandan Ravikiran, Rohit Saluja

Paper ↗
Aug 2025
arXiv 2025Text

The Art of Breaking Words: Rethinking Multilingual Tokenizer Design

Aamod Thakur, Ajay Nagpal, Atharva Savarkar and 7 others

Summary and full author list

Proposes improved pre-tokenisation for Indic scripts, reducing token-to-word ratios by ~6% and improving multilingual model efficiency.

Aamod Thakur, Ajay Nagpal, Atharva Savarkar, Kundeshwar Pundalik, Siddhesh Dosi, Piyush Sawarkar, Viraj Thakur, Rohit Saluja, Maunendra Sankar Desarkar, Ganesh Ramakrishnan

Paper ↗
Jul 2025
AIED 2025Text

CASSA: Context‑Aware Self‑Attention with Global Context Suppression and Relevance Modulation for MCQ Difficulty Estimation

Manikandan Ravikiran, Tarun Sharma, Arnav Bhavsar and 1 other

Summary and full author list

Proposes a self-attention mechanism with global context suppression and relevance modulation to better estimate the difficulty of multiple-choice questions in educational assessments.

Manikandan Ravikiran, Tarun Sharma, Arnav Bhavsar, Rohit Saluja

Paper ↗
Jul 2025
AIED 2025Vision

How Far Are We from Automatic Grading of Handwritten Cloze Form Questions?

Shrey Chandola, Manikandan Ravikiran, Rohit Saluja

Summary and full author list

Benchmarks current AI models against human graders on handwritten cloze-form examination responses, revealing significant gaps in automated grading accuracy and reliability.

Shrey Chandola, Manikandan Ravikiran, Rohit Saluja

Paper ↗
Jul 2025
arXiv 2025Speech

A2TTS: TTS for Low Resource Indian Languages

Ayush Singh Bhadoriya, Abhishek Nikunj Shinde, Isha Pandey and 1 other

Summary and full author list

A diffusion-based, speaker-conditioned TTS system for multiple Indian languages with zero-shot generation for unseen speakers.

Ayush Singh Bhadoriya, Abhishek Nikunj Shinde, Isha Pandey, Ganesh Ramakrishnan

Paper ↗
Jun 2025
arXiv 2025Vision

DRISHTIKON: Visual Grounding at Multiple Granularities in Documents

Badri Vishal Kasuba, Parag Chaudhuri, Ganesh Ramakrishnan

Summary and full author list

Locates specific text regions in complex multilingual document images at block, line, word, and point levels, with a new benchmark dataset.

Badri Vishal Kasuba, Parag Chaudhuri, Ganesh Ramakrishnan

Paper ↗
Jun 2025
arXiv 2025Text

Multilingual Tokenization through the Lens of Indian Languages: Challenges and Insights

N J Karthika, Maharaj Brahma, Rohit Saluja and 2 others

Summary and full author list

Shows that extremely low-resource Indian languages benefit significantly from tokenisers trained on related higher-resource languages, with a comprehensive evaluation across 17 Indic languages.

N J Karthika, Maharaj Brahma, Rohit Saluja, Ganesh Ramakrishnan, Maunendra Sankar Desarkar

Paper ↗
May 2025
ACL 2025Text

LexGen: Domain‑aware Multilingual Lexicon Generation

Ayush Maheshwari, Atul Kumar Singh, N J Karthika and 3 others

Summary and full author list

Creates multilingual dictionaries of domain-specific terms for Indian languages, enabling better coverage of technical vocabulary in low-resource settings.

Ayush Maheshwari, Atul Kumar Singh, N J Karthika, Krishnakant Bhatt, Preethi Jyothi, Ganesh Ramakrishnan

Paper ↗
May 2025
arXiv 2025Speech

Language Translation and Change of Accent for Speech‑to‑Speech Using Diffusion Models

Abhishek Mishra, Ritesh Sur Chowdhury, Vartul Bahuguna and 2 others

Summary and full author list

Reformulates speech translation and accent adaptation as a single conditional generation task, generating target-language speech with adapted accent from source phonemes.

Abhishek Mishra, Ritesh Sur Chowdhury, Vartul Bahuguna, Isha Pandey, Ganesh Ramakrishnan

Paper ↗
Apr 2025
TokShop @ ICML 2025Text

MorphTok: Morphologically Grounded Tokenization for Indian Languages

Maharaj Brahma, N J Karthika, Atul Singh and 5 others

Summary and full author list

Introduces morphology-aware preprocessing and Constrained BPE for Indic scripts, with a human evaluation metric called EvalTok.

Maharaj Brahma, N J Karthika, Atul Singh, Devaraj Adiga, Smruti Bhate, Ganesh Ramakrishnan, Rohit Saluja, Maunendra Sankar Desarkar

Paper ↗

Can't find a paper? Write to us at contact@bharatgen.com.