01
Political stance & framing
Study how models distinguish neutral reporting from politically framed content, without treating a legitimate viewpoint as misinformation.
AI Research · Software Engineering
Building AI that understands more languages.
I'm Madhusodan Chakraborty, a software engineer and researcher exploring trustworthy multilingual AI — with a focus on Bangla, political-stance evaluation, and reliable language technologies for underserved communities.
Multilingual language modelsInformation integrityRobustness & evaluation
01 / About
Research grounded in real-world software.
My work connects natural language processing with the engineering needed to build dependable systems. I am interested in how AI models interpret political language, identify misleading factual claims, and perform in languages that have fewer training and evaluation resources.
Alongside my research, I work on scalable backend services, data-processing pipelines, and enterprise applications. That practical perspective shapes how I think about model evaluation, reproducibility, and real-world use.
02 / Research direction
Making multilingual AI more dependable.
My research aims to identify where language models make mistakes, test methods to improve them, and create evaluation resources that other researchers and developers can use.
02
Evidence-based claim analysis
Investigate how language models assess verifiable factual claims and communicate uncertainty when credible evidence is missing.
03
Bias & robustness evaluation
Test counterfactual examples and adversarial paraphrases to understand whether models rely on irrelevant cues or fragile wording.
03 / Selected publication
Published research.
2025 · ACL Anthology · Bangla Language Processing Workshop
Read Between the Lines: A Benchmark for Uncovering Political Bias in Bangla News Articles
Coauthored research evaluating how large language models classify government-leaning, government-critical, and neutral Bangla news reporting. The work provides a foundation for further research into multilingual model reliability and political-stance evaluation.
View paper on ACL Anthology28LLMs evaluated
200Bangla news articles
Transformer fine-tuningContrastive learningCounterfactual augmentationAdversarial paraphrasingRetrieval-based verificationHuman-annotated datasets
04 / Engineering experience
From research ideas to dependable systems.
My background in enterprise software and language technologies helps me approach research with both experimental rigor and practical implementation in mind.
Backend software development
Java, Spring Boot, APIs, distributed services, and scalable application development.
Data pipelines and integration
ETL workflows, SQL databases, batch processing, and large-scale information management.
Bangla language technologies
Experience in Bangla information retrieval and language-processing research, including earlier work involving the Pipilika search engine.
05 / Connect
Interested in trustworthy multilingual AI?
I'm interested in research discussions and collaborations around language-model evaluation, low-resource languages, and dependable information systems. Explore my research and publications below.
View Google Scholar profile