Shariar Kabir

Shariar Kabir

AI Safety, Interpretability, Agentic Systems

Publications

You can also find my articles on my Google Scholar profile.

Conference Papers

Circuit Oracle: Automating Attribution Graph Analysis via Natural-Language Queries

Published in MechInterp & CompLearn Workshops @ ICML 2026, 2026

A multi-agent system that automates mechanistic-interpretability analysis by autonomously answering natural-language questions about a model through multi-hop traversal of its attribution graph.

Recommended citation: Hong Kiat Tan, Shariar Kabir, Swastik Agrawal, Sai Chereddy, Sriram Balasubramanian. (2026). "Circuit Oracle: Automating Attribution Graph Analysis via Natural-Language Queries." MechInterp & CompLearn Workshops at ICML 2026.

PReSS: A Black-Box Framework for Evaluating Political Stance Stability in LLMs via Argumentative Pressure

Published in PoliticalNLP @ LREC-COLING 2026, 2026

A black-box framework that stress-tests LLMs’ stated opinions using argumentative challenges, classifying responses as stable or unstable across political topics.

Recommended citation: Shariar Kabir, Kevin Esterling, Yue Dong. (2026). "PReSS: A Black-Box Framework for Evaluating Political Stance Stability in LLMs via Argumentative Pressure." PoliticalNLP @ LREC-COLING 2026.
Download Paper

SynthNID: Synthetic Data to Improve End-to-end Bangla Document Key Information Extraction

Published in BLP Workshop at EMNLP 2023, Singapore, 2023

A system to generate domain-specific synthetic document image data for training OCR-less end-to-end Key Information Extraction systems for low-resource languages.

Recommended citation: Syed Monsur, Shariar Kabir, Sakib Chowdhury. (2023). "SynthNID: Synthetic Data to Improve End-to-end Bangla Document Key Information Extraction." BLP Workshop at EMNLP 2023, pages 117–123, Singapore.
Download Paper

Journal Articles

Preprints

AmarDoctor: An AI-Driven, Multilingual, Voice-Interactive Digital Health Application

Published in arXiv preprint arXiv:2510.24724, 2025

A multilingual voice-interactive digital health app providing patient triage and AI-driven clinical decision support for Bengali speakers.

Recommended citation: Nazmun Nahar, Ritesh Harshad Ruparel, Shariar Kabir, Sumaiya Tasnia Khan, Shyamasree Saha, Mamunur Rashid. (2025). "AmarDoctor: An AI-Driven, Multilingual, Voice-Interactive Digital Health Application." arXiv preprint arXiv:2510.24724.
Download Paper

When Models Refuse: Political Steerability and Feature Richness as Measures of Ideological Depth

Published in arXiv preprint arXiv:2508.21448, 2025

Defines ideological depth as a model’s steerability and the feature richness of its internal political representations measured with sparse autoencoders.

Recommended citation: Shariar Kabir, Kevin Esterling, Yue Dong. (2025). "When Models Refuse: Political Steerability and Feature Richness as Measures of Ideological Depth." arXiv preprint arXiv:2508.21448.
Download Paper

Automatic Speech Recognition for Biomedical Data in Bengali Language

Published in arXiv preprint arXiv:2406.12931, 2024

A prototype ASR system for Bengali biomedical data, covering Bengali and Sylheti dialects, trained on a 46-hour domain-specific corpus.

Recommended citation: Shariar Kabir, Nazmun Nahar, Shyamasree Saha, Mamunur Rashid. (2024). "Automatic Speech Recognition for Biomedical Data in Bengali Language." arXiv preprint arXiv:2406.12931.
Download Paper