Publications, talks, posters, and academic work in AI, ML, and NLP.
Technical reports
ConceptGate: Learning and Steering Concepts in Language Models [2026]
A few-shot, training-free adapter that detects a concept from a frozen model’s own layers and steers generation along a closely related direction, with interactive figures over real GPT-2, Qwen2.5-0.5B and gemma-2-2b runs.
[Read]
Response Provenance: Tracing Agent Claims Back to Their Causes [2026]
An agent writes a paragraph. Which sentence came from which file, which instruction, which earlier message? A practical method for answering that on hosted APIs, where every state-of-the-art attribution method is unavailable by construction.
[Read]
Publications
| *Google Scholar: 67 citations | h-index: 3 | i10-index: 3 — as of August 2026* |
INDUS-SDE [KDD 2026]
N. Pantha, S. Awale, V. Kuruvanthodi, S. KC, M. Ramasubramanian, C. Davis, B. Praveen, E. Foshee, B. Bhattacharjee, K. Bugbee, R. Ramachandran, “INDUS-SDE: A Language Model for Scientific Content Curation and Discovery,” in Proc. 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ‘26), V.2, AI for Sciences Track, pp. 11762–11773, Jeju Island, Republic of Korea, Aug. 2026. [DOI] [Blog]
A small encoder language model for scientific content curation and discovery, designed for sparse contexts where scientific terminology is rare but carries most of the signal. Deployed in NASA’s Science Discovery Engine pipeline.
Scientific Code Search at Scale [2026]
N. Pantha, P. R. Kumbam, S. Awale, P. Krishnappa, M. Ramasubramanian, N. Jha, E. Foshee, A. Kumar, R. A. Slank, A. Danehkar, R. Ramachandran, “Scientific Code Search at Scale: A Multi-Domain Dataset and Benchmark,” arXiv preprint arXiv:2607.05443, 2026. [arXiv]
A multi-domain dataset of 5,264 scientific repositories across five NASA SMD divisions, with two information-retrieval benchmarks — 219 expert-curated queries over 117,950 code snippets — for evaluating scientific code search.
INDUS [EMNLP 2024]
B. Bhattacharjee, A. Trivedi, M. Muraoka, M. Ramasubramanian, T. Udagawa, I. Gurung, N. Pantha, et al., “INDUS: Effective and Efficient Language Models for Scientific Applications,” EMNLP 2024, Industry Track, pp. 98–112, Miami, FL, 2024. [25 citations] [Paper] [DOI]
Domain-adapted language models for Earth science, biology, physics, heliophysics, planetary sciences and astrophysics.
Guardrailing LLMs for Science [2024]
N. Pantha, M. Ramasubramanian, I. Gurung, M. Maskey, R. Ramachandran, “Challenges in Guardrailing Large Language Models for Science,” arXiv preprint arXiv:2411.08181, 2024. [21 citations] [arXiv]
A taxonomic framework for LLM guardrails in scientific applications.
Feature Selection [IEEE SoutheastCon 2024]
N. Pantha, M. Ramasubramanian, I. Gurung, M. Maskey, L. M. Sanders, et al., “Feature Selection in High-Dimensional Space with Applications to Gene Expression Data,” IEEE SoutheastCon 2024, Atlanta, GA, pp. 6–15, 2024. [3 citations] [DOI]
Blaze [IEEE CLOUD 2023]
S. Marru, B. Freitag, D. Wannipurage, N. Pantha, et al., “Blaze: A High-Performance, Scalable, and Efficient Data Transfer Framework,” IEEE 16th International Conference on Cloud Computing (CLOUD), pp. 58–68, 2023. [11 citations]
Evaluated transatlantic data transfer between ESA and NASA.
Reasoning Risks Benchmark [NeurIPS 2026, under review]
T. Tchrakian, A. Pascale, N. Pantha, J. Barry, N. Jha, R. A. Slank, A. Danehkar, E. Foshee, G. De Mel, M. Ramasubramanian, J. Carnerero-Cano, R. Ramachandran, J. Bernabe-Moreno, “A Benchmark for Reasoning Risks in Scientific Content Generation,” submitted to NeurIPS 2026, Evaluations and Datasets Track.
IEEE GRSS Workshop Report [2022]
D. That, N. Pantha, et al., “Report on the IEEE GRSS Workshop on Remote Sensing Data Management Technologies in Geoscience 2022 [Technical Committees],” IEEE Geoscience and Remote Sensing Magazine, vol. 10, no. 4, pp. 273–277, Dec. 2022. [DOI]
Query by Humming [IJASS 2019]
P. Koirala, M. Chapagain, N. Pantha, N. B. Adhikari, “Effects of Auto Tuning and Pitch Normalization on Query by Humming,” International Journal of Advanced Social Sciences (IJASS), vol. 1, no. 2, pp. 11–16, 2019. [1 citation] [PDF]
ABCDE [2017]
N. Pantha, K. Mandal, A. Parajuli, “Artificial Intelligence, Big Data and Cloud Driven E-Governance (ABCDE),” 2017. [DOI]
Talks
Context & Harness Engineering: Lessons from AKD [May 2026]
Location: Huntsville, AL — 2nd ESA-NASA Workshop on AI Foundation Model for Earth Observation
On what it actually takes to engineer context and harnesses for agentic systems, drawn from building Accelerated Knowledge Discovery (AKD) at NASA-IMPACT. Delivered in the “Building a GeoAI Agent: A Hands-On Tutorial on Agentic Foundation Models for Earth Observation” session, which I co-organized and taught.
Coverage: UAH ESSC News
LLMs and Agentic Workflows in the Geospatial Domain [April 2024]
Location: IEEE GRSS Hackathon, Chennai, India
A talk on applying large language models and agentic workflows to geospatial problems, given as an organizer and evaluator of the hackathon.
Evaluating LLMs for Open Science RAG Systems [March 2024]
Location: NASA SMD AI Workshop
On evaluation strategy for retrieval-augmented generation systems built for open science. Related material contributed to the LLM Cookbook for Open Science.
Engineering-AI [May-July, 2019]
Location: JEC, Bhaktapur, Nepal
As a part-time lecturer for teaching the engineering course AI at JEC, I cover the whole course through the expected timeline.
Slide: nish1001.github.io/engineering-ai/
AI Use-Cases [March 26, 2019]
Location: Amman, Jordan
The talk is a part of presentation that our COMPANY Mpercept Tech gave in Amman, Jordan to a telecommunication. This part of presentation is where I talk about general introduction to Artificial Intelligence and different usecases for systems that can actually implement AI.
Slide: nish1001.github.io/ai-usecases
Debunking AI [May 04, 2018]
Location: Kaffe Codes, Thapathali, Kathmandu, Nepal
The talk is about debunking the myths and realities of AI and hype. It was given at DMH Friday Session 10 organized by Swopna Digital at Kaffe Codes.
Slide: nish1001.github.io/debunking-ai
Let’s Kickstart ML [October 14, 2017]
Location: Patan College, Kupondole, Lalitpur, Nepal
The talk is about how any person can kickstart their path to AI and Machine Learning. It was given at first AI meetup in Kathmandu.
Slide: nish1001.github.io/lets-kickstart-ml
Video: youtube.com/watch?v=s0jfB4Ps9O0
Writing Clean Code [August 19, 2017]
Location: Fusemachines Nepal, Hattisar, Kathmandu, Nepal
This is somewhat agnostic talk about how we can render a code clean. It was given in Python Nepal Meetup #12.
Slide: nish1001.github.io/writing-clean-code
Posters & conference presentations
GeoGuard [ESA-NASA Workshop 2026]
N. Pantha, R. Sahoo, S. Thapa, M. Ramasubramanian, R. Ramachandran, “GeoGuard: An Agentic Guardrails and Validation Framework for Geospatial AI,” 2nd ESA-NASA Workshop on AI Foundation Model for Earth Observation (poster), Huntsville, AL, May 2026. [Workshop] [Coverage]
An agentic verification system that checks the outputs of GeoAI systems.
GeoUI Intent Protocol [ESA-NASA Workshop 2026]
S. Thapa, N. Pantha, G. Panthee, R. Sahoo, M. Ramasubramanian, I. Gurung, R. Ramachandran, “GeoUI Intent Protocol: A Protocol Between AI Agents and Geospatial Visualization Tools,” 2nd ESA-NASA Workshop on AI Foundation Model for Earth Observation (poster), Huntsville, AL, May 2026. [NTRS]
AKD Labs [ESA-NASA Workshop 2026]
G. Panthee, N. Pantha, S. Thapa, S. Awale, S. KC, P. Krishnappa, M. Ramasubramanian, R. Ramachandran, “AKD Labs: An End-to-End Platform for Designing, Building, Debugging, and Benchmarking AI Agents,” 2nd ESA-NASA Workshop on AI Foundation Model for Earth Observation (poster), Huntsville, AL, May 2026. [Workshop]
Extending INDUS for Sentence Embeddings [AGU 2025]
V. Kuruvanthodi, B. Bhattacharjee, M. Elkaref, G. R. De Mel, N. Pantha, et al., “Extending INDUS Models for Low-Latency, High-Context Scientific Sentence Embeddings,” AGU 2025 Fall Meeting, 2025.
Automation Pipeline for ML-Assisted Curation [2025]
N. Pantha, B. Praveen, M. Ramasubramanian, S. Awale, S. KC, “Automation Pipeline for ML-Assisted Scientific Data Curation and Discovery,” 2025 Jamboree, Marshall 65 Spring Showcase, Huntsville, AL, May 2025. [NTRS]
NLP Applications of Domain-Adapted Language Models [AGU 2024]
M. Ramasubramanian, I. Gurung, N. Pantha, et al., “NLP Applications of Domain-Adapted Language Model for Enhanced Scientific Data Discovery,” AGU 2024 (IN52A-08, poster), Washington, DC, Dec. 2024. [AGU]
INDUS [EMNLP 2024]
B. Bhattacharjee, et al., N. Pantha, et al., “INDUS: Effective and Efficient Language Models for Scientific Applications,” EMNLP 2024, Industry Track (poster), Miami, FL, Nov. 2024.
Evaluation of NASA SMD Large Language Model [AGU 2023]
M. Ramasubramanian, N. Pantha, I. Gurung, et al., “Evaluation of NASA Science Mission Directorate Large Language Model,” AGU 2023 (IN24A-01, oral presentation), San Francisco, CA, Dec. 2023. [AGU]
Leveraging LLMs for NASA Search and Discovery [AGU 2023]
A. Acharya, I. Gurung, A. Bhusal, C. E. Phillips, C. Davis, M. Ramasubramanian, R. Ramachandran, N. Pantha, et al., “Leveraging Large Language Models to Enhance NASA’s Information Search and Discovery,” AGU 2023 (IN53A-05), San Francisco, CA, Dec. 2023.
ML Benchmark for Space-Flown Rodent RNA-Seq [AGU 2022]
N. Pantha, et al., “Machine Learning Based Benchmark for Space Flown Rodent Liver RNA Sequencing Data in Space Biology Research,” AGU 2022 (IN22D-0332, poster), Chicago, IL, Dec. 2022. [iPoster] [NTRS]
Technical articles & media
INDUS-SDE on the NASA Science Data Blog [2026]
NASA IMPACT AI Team, “INDUS-SDE: A Language Model for Scientific Content Curation and Discovery,” NASA Science Data Blog, Sep. 2026. [Article]
Coverage of INDUS-SDE and its deployment in the Science Discovery Engine pipeline.
UAH Researchers at the ESA-NASA AI Workshop [2026]
“UAH Lab for Applied Science Researchers Engage in International AI Workshop,” UAH Earth System Science Center News, Aug. 2026. [Article]
Coverage of the 2nd ESA-NASA Workshop on AI Foundation Model for Earth Observation, featuring my poster work on evaluation and benchmarking for geospatial foundation models.
Revolutionizing Scientific Discovery with AI [2025]
N. Pantha, M. Ramasubramanian, C. Davis, D. Koehl, “Revolutionizing Scientific Discovery with AI: Inside the Science Discovery Engine,” NASA Science Discovery Engine Blog, May 2025. [Article]
Of Mice and Machines [2023]
NASA (feat. N. Pantha), “Of Mice and Machines: Using Machine Learning to Study Space Radiation in Mice,” NASA Earthdata Blog, Dec. 2023. [Article]