Large language models & LLM evaluation
How commercial LLMs respond to framing, instruction and manipulation — and how to measure that reliably.
Research areas
Multilingual NLP & LLM bias
Whether a model's behaviour holds up outside English. Doctoral work across English, Hindi and Bengali.
Generative & agentic AI
Applied GenAI, RAG and agentic systems delivered for healthcare and enterprise clients at Accenture.
Social-media threat detection
Predicting radicalization and civil-unrest risk from public text before violence forms.
Health & wellbeing NLP
Mental-health screening signals and drug–disease prediction from unstructured text.
Opinion mining & sentiment
How public sentiment moves, and what people actually worry about, measured at scale.
Anomaly & outlier detection
Density-based algorithms that stay honest when the data itself varies in density.
Information retrieval & topic modelling
Finding and summarising what matters in large text collections — taught and researched at IEM Kolkata, and patented for offline use on a single device.
Current doctoral research
Bias detection and mitigation in multilingual large language models
IIIT Kalyani, since August 2022 — part-time alongside full-time industry work. Supervised by Dr. Imon Mukherjee, co-supervised by Dr. Debarshi Kumar Sanyal (IACS, Kolkata).
The central question is whether a model's fairness holds up outside English. The flagship study measured cognitive-bias vulnerability across English हिन्दी বাংলা in six commercial LLMs over 683,616 responses, and found confirmation-pattern prompts cut neutral answers by 19.42 percentage points on average — with vulnerability higher in Hindi and Bengali.
Read about this work →Research papers
2026 · Journal article
How prompts manipulate models: measuring cognitive bias vulnerability in commercial LLMs across languages
Innovations in Systems and Software Engineering (Springer)
Read the explainer → 2022 · Conference paperStudying Borderline Personality Disorder Using Machine Learning
16th IEEE International Conference on Ubiquitous Information Management and Communication (IMCOM), Seoul
Read the explainer → 2022 · Book chapterCrime by Crowd is Not Crime any More: A Predictive Analysis of Riot and Radicalization Oriented Text of Social Media
Applications of Machine Intelligence in Engineering (CRC Press)
Read the explainer → 2022 · Book chapterAssessing the Concerns of People in the Pandemic Situation: An Analytical Approach on the Social Media Data
Applications of Machine Intelligence in Engineering (CRC Press)
Read the explainer → 2021 · PreprintA Comparative Study on Credit Card Fraud Detection
Engineering Archive (engrXiv) preprint
Read the explainer → 2021 · Journal articleDisease Prediction from Drug Information using Machine Learning
American Journal of Electronics & Communication / engrXiv
Read the explainer → 2021 · Journal articleDoes Social Media Follow News Media? A Comparative Sentiment Analysis During the COVID-19 Pandemic
International Journal of Information Communication Technologies and Human Development 13(4), IGI Global
Read the explainer → 2019 · Book chapterA framework for predicting and identifying radicalization and civil unrest oriented threats from WhatsApp group
Emerging Technology in Modelling and Graphics, AISC vol. 937 (Springer)
Read the explainer → 2018 · Conference paperA Relative Kernel-Density Based Outlier Detection Algorithm (RKDOS)
12th IEEE International Conference on Software, Knowledge, Information Management and Applications (SKIMA), Phnom Penh
Read the explainer →Patents
2019 · Patent 201931009037
📱 Offline multilingual NLP device
Language AI mostly lives in the cloud: without a connection there is no intelligence, and every query leaves the device.
Read the explainer → 2019 · Patent 201931053496🔥 IoT fire detection & robot-assisted evacuation
In a building fire, detection is only half the problem — most casualties happen during the evacuation itself.
Read the explainer → 2024 · Patent 202431013004🚁 Chemical-free drone pest control
Pesticides kill pests along with pollinators and soil life, while resistance keeps rising and regulators keep withdrawing compounds.
Read the explainer →