Research faculty at Virginia Tech
I build and evaluate generative AI, LLM, vision-language, computer vision, and machine learning systems for real-world problems in public health, intelligent transportation, digital content understanding, and AI for social good.
About me
I am a research faculty member at Virginia Tech in Blacksburg, USA, building and evaluating generative AI, large language model, vision-language, and multimodal machine learning systems for high-impact real-world settings. My recent work focuses on LLM evaluation, agentic AI, RAG, multilingual and low-resource reasoning, benchmark construction, model bias, and responsible AI deployment.
I apply this work across health AI, online safety, and intelligent transportation, with projects in non-invasive health monitoring, misinformation and harmful-content detection, driver behavior analysis, ADS readiness, and interpretable multimodal perception.
Research areas
My research centers on large language models, multimodal learning, and deep learning systems that can be evaluated, explained, and deployed responsibly in health care, transportation, and online safety settings.
I work on LLM evaluation, multilingual and low-resource reasoning, RAG and retrieval-augmented distillation, agentic AI frameworks, prompt-based learning, model bias, and responsible deployment of generative AI in high-impact domains.
Non-invasive cardiovascular signal recovery, biomedical vision-language models, LLM-assisted health applications, mental health surveillance, medical dialogue generation, and AI methods for neurodegenerative disease modeling.
LLM- and VLM-powered systems for online harms, misinformation, hate speech, polarization, mental health, code-mixed content, low-resource languages, and interpretable multimodal content moderation.
Large vision-language models for traffic-scene understanding, driver behavior analysis, naturalistic driving data, face de-identification, ADS readiness, 3D perception, and wearable sensing for stress and fatigue.
Tools and technologies
I work across the full applied-AI stack: building models, designing evaluations, engineering data pipelines, containerizing workflows, running cloud experiments, and translating research ideas into usable systems.
Hands-on work with pretraining, fine-tuning, RAG, LangChain, RLHF, DPO, evaluation, bias analysis, prompt-based learning, and agentic workflows.
Implementation-heavy work across deep learning, computer vision, vision-language models, signal processing, detection, segmentation, tracking, and explainability.
Python-first research engineering with scalable pipelines for data processing, experimentation, visualization, cloud execution, and deployment.
Selected publications
A selected view of my work across language, health, transportation, and multimodal AI. The full publication list is available on Google Scholar.
Online safety, polarization, low-resource language understanding, and multimodal social content.
Selected papers, newest firstVargas F., Trager J., Alves D., Guida M., Thapa S., Atil B., Dementieva D., Smart A.J., and Agrawal A.
Naseem U., Thapa S., and collaborators.
Shiwakoti S., Thapa S., Rauniyar K., Maharjan I., and Naseem U.
Thapa S., Veeramani H., Hu L., Zhang Q., Wang W., and Naseem U.
Naseem U., Shiwakoti S., Shah S., Thapa S., and Zhang Q.
Thapa S., Rauniyar K., Veeramani H., Adhikari S., Razzak I., and Naseem U.
Thapa S., Rauniyar K., Shiwakoti S., Adhikari S., Rashid J., Kim J., and Naseem U.
Shah S., Thapa S., Jain A., and Huang L.
Biomedical NLP, non-contact physiological sensing, and AI for mental and neurological health.
Selected papers, newest firstThapa S., Rauniyar K., Veeramani H., Shah A., Razzak I., and Naseem U.
Naseem U., Thapa S., and Masood A.
Naseem U., Thapa S., Zhang Q., Hu L., and Nasim M.
Deshpande Y., Thapa S., Sarkar A., and Abbott A.L.
Li F., Thapa S., Bhat S., Sarkar A., and Abbott A.L.
Thapa S., Salman M., Shah S., Zhang Q., Rashid J., Hu L., Razzak I., and Naseem U.
Thapa S., Salman M., Shah S., Shiwakoti S., Zhang Q., Hu L., Razzak I., and Naseem U.
Adhikari S., Thapa S., Naseem U., Singh P., Huo H., Bharathy G., and Prasad M.
Thapa S., Adhikari S., Ghimire A., and Aditya A.
Traffic-scene understanding, driver privacy, naturalistic driving, 3D perception, and driver health.
Selected papers, newest firstGuduri B., Costa R., and Thapa S.
Thapa S. and Sarkar A.
Thapa S. and Sarkar A.
Jain S., Thapa S., Chen K.T., Abbott A.L., and Sarkar A.
Jain S., Thapa S., Bhardwaj S., Abbott A.L., Sarkar A., and Xuan J.
Thapa S. and Sarkar A.
Thapa S. and Sarkar A.
Thapa S., Cook J., and Sarkar A.
Experience
Research Faculty
Graduate Research Assistant
Worked on transportation AI, naturalistic driving data, DOT/National Academies research synthesis, and driver face de-identification.
Graduate Research Assistant
Developed non-contact cardiovascular sensing methods for blood-volume-pulse recovery from face video.
Visiting Scholar
Built AI models for early Alzheimer's disease detection using neuropsychological, brain-volume, and speech-transcript signals.
Openlab Summer Intern
Worked on conditional progressive GANs for spectrally valid satellite image generation and validation with UNOSAT water-stream detection models.
M.S., Computer Science (Thesis); Graduate Certificate in Data Analytics
Blacksburg, Virginia.
B.Tech., Software Engineering
New Delhi, India.
Service
I serve the research community through area chair roles, program committees, reviewing, workshop organization, and shared-task leadership across AI, NLP, social computing, multimedia, vision, and intelligent systems venues.
I have contributed as an Area Chair for major venues such as ACM Multimedia, ICASSP, and IJCNN. These roles involve coordinating expert review, evaluating technical contributions, and helping shape strong conference programs in multimedia, neural networks, signal processing, and AI.
I serve as a program committee member and reviewer for major AI, NLP, and computer vision venues, including NeurIPS, ICLR, CVPR, ECCV, ICCV, ACL, EMNLP, ARR, AAAI, SIGIR, WebConf, ICWSM, LREC, COLING, ECAI, IJCAI, ECML/PKDD, ACM CSCW, and multiple IEEE/ACM transactions and conferences.
I have chaired and organized a broad set of workshops that bring together researchers working on low-resource NLP, multimodal social-good AI, biomedical vision-language modeling, online safety, and socio-political event extraction.
Selected roles, newest first
I help lead shared tasks that define benchmark datasets, evaluation protocols, and community research agendas for multilingual NLP, multimodal content understanding, online safety, polarization, misinformation, and AI for healthier digital spaces.
Selected tasks, newest first