ML Researcher
I build multimodal and agentic AI, systems that see, read, and act by reasoning across text, images, and video.
I recently finished my Master's in Computer Science at Arizona State University, advised by Yezhou Yang, where I still collaborate with the lab. My work spans vision-language models, preference optimization for diffusion models, and agentic systems that turn perception into action. Before ASU, I earned a B.E. from KLE Technological University (2023), advised by Shankar Gangisetty on visual question answering.
@article{hegde2025chartqa,
title={ChartQA-X: Generating Explanations for Charts},
author={Hegde, Shamanthak and Fazli, Pooyan and Seifi, Hasti},
journal={arXiv preprint arXiv:2504.13275}, year={2025}}
@article{saeidi2025dual,
title={Dual Caption Preference Optimization for Diffusion Models},
author={Amir Saeidi and Yiran Lawrence Luo and Agneet Chatterjee and Shamanthak Hegde and Bimsara Pathiraja and Yezhou Yang and Chitta Baral},
journal={Transactions on Machine Learning Research}, year={2025}}
@InProceedings{Verma_2024_CVPR,
author = {Verma, Aayush Atul and Saeidi, Amir and Hegde, Shamanthak and Therala, Ajay and Bardoliya, Fenil Denish and Machavarapu, Nagaraju and Ravindhiran, Shri Ajay Kumar and Malyala, Srija and Chatterjee, Agneet and Yang, Yezhou and Baral, Chitta},
title = {Evaluating Multimodal Large Language Models Across Distribution Shifts and Augmentations},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June}, year = {2024}, pages = {5314-5324}}
@inproceedings{hegde2023making,
title={Making the V in Text-VQA Matter},
author={Hegde, Shamanthak and Jahagirdar, Soumya and Gangisetty, Shankar},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={5580--5588}, year={2023}}
@inproceedings{alampalle2023weakly,
title={Weakly Supervised Visual Question Answer Generation},
author={Alampalle, Charani and Hegde, Shamanthak and Jahagirdar, Soumya and Gangisetty, Shankar},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={5589--5597}, year={2023}}
A reverse-proxy that intercepts live HTTPS traffic and runs a local LLM to flag and redact healthcare identifiers (MRNs, SSNs, ICD-10 codes, medication names) before any request reaches an external API. Ships with a compliance dashboard and automated voice alerts.
An agentic system (Gemini + LangChain + Playwright) that pulls structured fields out of dynamic pages and watches them for changes, running on a serverless FastAPI + AWS Lambda backend.
Training and evaluating multimodal LLMs and VLMs (SFT, DPO/GRPO) for robustness benchmarking and diffusion-model alignment. Work published at TMLR and CVPR workshops.
Built a unified vision-language framework for chart question answering and explanation generation, published as ChartQA-X (WACV 2026).
Resolved production bugs and integrated IronPython scripting into DeviceBridge, a distributed industrial data platform.
