AI, LLMs and research
Multilingual LLM bias: ask the same question in Bengali
Manipulative prompts reduced neutral answers; vulnerability was higher in Hindi and Bengali than in English.
RAG explained for teachers using their own documents
A plain look at why models invent facts and how retrieval tries to ground them in your documents.
Evaluating LLMs at work: a practical test routine
Test five real cases before you hand any model your work.
AI for teachers
24 practical guides. See them all →
Generative AI for teachers: a plain explanation that helps
It predicts the next words from patterns, so fluent output is not the same as correct output, and everything it produces gets checked.
How teachers can use AI: five jobs, one decision
Five verbs, one boundary, and the prompt that starts small tonight.
AI guide for teachers: a practical first week
One low-risk task a day, ten minutes each, nothing reaches students until you read every word.
Generative AI in education: a department plan
Drafting moves to the tool. Judgment stays with the teacher.
AI tools for teachers: the three kinds worth knowing
Pick the right kind for the job in front of you and keep the rest simple.
AI for lesson planning without losing the lesson
The tool drafts the period. You check it and keep control.
AI for classroom activities: getting past the worksheet
The tool suggests the activity. You check it against your objective, your room, and your students.
AI for teacher feedback that still sounds like you
Spot patterns across a class set, build a reusable comment bank, and keep every word that reaches students in your own voice.
AI for assessment: Questions that reveal thinking
Draft the items. Check them. Decide what reaches your students.
Research explainers
Every paper and patent, explained without the jargon.
How prompts manipulate models
Plain-language explainer: how manipulative prompt framing sways commercial LLMs, measured across English, Hindi and Bengali over 683,616 responses.
Studying Borderline Personality Disorder Using Machine Learning
Plain-language explainer: detecting borderline personality disorder signals from emotional patterns in text, at around 81% accuracy.
Crime by Crowd is Not Crime any More
Plain-language explainer: estimating riot probability from radicalization-oriented text on social media before violence breaks out.
Assessing the Concerns of People in the Pandemic Situation
Plain-language explainer: LDA topic modelling over pandemic-era Reddit posts to surface what people actually worried about.
A Comparative Study on Credit Card Fraud Detection
Plain-language explainer: comparing unsupervised methods — K-means, PCA, t-SNE and Self-Organising Maps — for surfacing fraudulent card behaviour as outliers.
Disease Prediction from Drug Information using Machine Learning
Plain-language explainer: predicting which disease a drug treats from 215,063 drug reviews, with Random Forest reaching 0.83 accuracy.
Predicting Radicalization Threats from WhatsApp Groups
Plain-language explainer: a framework combining NLP, semi-supervised learning and topic detection to score WhatsApp groups for civil-unrest threat.
A Relative Kernel-Density Based Outlier Detection Algorithm
Plain-language explainer: RKDOS scores outliers by relative kernel density using adaptive Gaussian kernels over k-nearest and reverse-nearest neighbours.
Does Social Media Follow News Media?
Plain-language explainer: comparing news and Twitter sentiment (Jan–Jul 2020) with TextBlob, VADER and IBM Watson — news negativity peaked first.
Offline NLP Device — Patent
Granted Indian patent: a single device that performs real-time NLP — summarisation, topic detection, query-driven retrieval — over text, audio, video and images, entirely
IoT Fire Detection & Evacuation — Patent
Granted Indian patent: an IoT sensor network with two-level control that detects fire, texts dynamically updated escape routes to occupants, and deploys firefighting robo
Drone-Augmented Pest Control — Patent
Indian patent: a camera-equipped drone swarm that detects crop pests with SSD deep learning and neutralises them with targeted magnetron pulses — no chemicals.