A deep learning system that classifies skin conditions including acne, eczema, psoriasis, and melanoma from patient-uploaded photographs. The model uses CNN architectures trained on labelled dermatology datasets, helping rural clinics and health apps provide preliminary screening without a specialist.
Uses MediaPipe face and pose landmark tracking to detect early neurological warning signs of stroke — facial drooping, asymmetric eye movement, and abnormal posture — through real-time webcam analysis. Designed for assisted living environments and emergency screening.
A hybrid fraud detection engine combining traditional ML classifiers with deep learning anomaly detection to flag suspicious financial transactions in real time. Processes transaction streams, computes risk scores, and automatically suspends flagged accounts pending review.
Constructs a personalised digital simulation of a patient's brain tumor using MRI scan data, predicting tumor growth trajectories and simulating treatment responses. Combines segmentation models with physics-informed neural networks to assist oncologists in treatment planning.
A real-time accessibility tool that captures hand gestures via webcam, recognises American/Indian Sign Language alphabets and words using a CNN/LSTM pipeline, and converts them to text and synthesised speech — enabling communication for the hearing/speech impaired.
A unified ML-based diagnostic platform that predicts multiple chronic conditions — diabetes, heart disease, liver disease, kidney disease — from patient health metrics entered through a web form. Each disease uses a separately trained and optimised model with explainability output.
An NLP model that analyses text from social media posts, captions, and written responses to predict Big Five personality traits (OCEAN model). Applications include HR screening, personalised marketing, and mental wellness apps.
Monitors electrical grid sensor data streams in real time to detect faults, energy theft, and abnormal consumption patterns using time-series anomaly detection models. Designed for smart city and industrial energy management deployments.
A privacy-preserving intrusion detection system where multiple network nodes each train a local ML model and share only model weights — not raw data — with a central aggregator. XAI (LIME/SHAP) explains which features triggered each alert.
An ML system that predicts early-stage Alzheimer's disease from MRI-derived features, cognitive test scores, and biomarker data. Combines traditional ML classifiers with deep learning image analysis to produce a risk score with confidence intervals.
Farmers upload photos of crop leaves, and a CNN model identifies diseases like blight, rust, and mildew with confidence scores. The system also provides treatment recommendations and tracks disease history per farm.
A webcam-based emotion recognition system that classifies seven human emotions (happy, sad, angry, surprised, fearful, disgusted, neutral) in real time using a CNN trained on facial expression datasets.
An HR tool that accepts a job description and a batch of uploaded resumes, then uses NLP (TF-IDF + BERT similarity) to rank candidates by relevance, extract key skills, and generate a shortlist report — automating the first stage of recruitment.
A time-series deep learning system that forecasts stock prices using LSTM networks trained on historical price data combined with sentiment scores extracted from financial news headlines via NLP.
A web application where users paste or upload news articles, and an NLP pipeline (TF-IDF + ensemble classifier) predicts whether the content is real or fabricated — with a confidence score and explanation of key triggering terms.
An intelligent FAQ chatbot for college websites that answers student queries about admissions, exams, fees, and campus life using a retrieval-based NLP model. Escalates unresolved queries to a human admin.
A real-time safety system that monitors a driver's eye blink rate and yawn frequency through a webcam feed, calculating an Eye Aspect Ratio (EAR) to detect drowsiness and trigger audio-visual alerts before an accident can occur.
Predicts crop yield based on soil composition parameters, historical weather data, rainfall, temperature, and fertiliser usage using regression models. Provides district-level advisory recommendations to farmers and agricultural officers.
ADR Lab provides complete development — code, documentation, deployment, and viva support.