Projects

Deep learning system for automated stenosis detection from coronary angiograms. Implements CNN with attention mechanisms for medical image analysis in low-resource settings.

User-independent deep learning framework for assistive wheelchair navigation using surface electromyography signals. Real-time gesture recognition without user calibration.

Two-tier deep learning framework for real-time vehicle classification from acoustic horn signatures. Privacy-preserving traffic monitoring for smart city applications.

A comprehensive digital platform bridging the gap between citizens and law enforcement in Bangladesh. Features AI-driven fake report detection using CNN-based sentiment analysis, anonymous reporting with encrypted data handling, Emergency SOS with geo-fencing, and real-time case tracking for transparency and accountability.

A domain-adapted AI assistant providing reliable academic guidance for Bangladeshi students. Integrates Retrieval-Augmented Generation (RAG) with NCTB textbooks and official university sources to reduce hallucinations. Features two-tier LLM architecture, bilingual support (Bangla/English), and automated web crawling for study abroad data. The NCTB Class-8 Science & BGS system, a subject-adaptive eight-layer RAG pipeline, is accepted at iCACCESS 2026.

A dual-platform system built for students and companies to simplify the hiring process. Students can apply to thousands of jobs with one click, while an AI engine matches their profiles to the best-fit positions. Companies can auto-invite qualified candidates for interviews and conduct online exams directly on the platform, reducing manual work by up to 90%.

An interactive health assistant where users describe their symptoms and the chatbot responds like a doctor to predict possible diseases. Built using a rule-based AI with word matching and an LLM backend, it supports symptom analysis, disease prediction, and conversation memory for follow-up queries.

A full-stack Flutter marketplace modernizing Bangladesh's wedding and event industry. Features an automated Bangla Budget Smart Calculator that allocates funds across venue, attire, decor, and photography pillars, plus a Blind Reverse-Bidding Engine where vendors submit sealed bids without seeing competitors' prices. A 7-slot curated matching algorithm ensures emerging freelancers get equal visibility alongside legacy agencies.

This website - a static research portfolio built with modern web standards, featuring dark mode design, responsive layout, and semantic HTML structure.

A browser-native platform that turns any web browser into a personalized AI speech-language pathologist. It helps people with speech impediments, stuttering, and stroke-induced vocal loss through gamified, voice-interactive exercises. The core is an Adaptive Reinforcement Learning Engine that dynamically customizes recovery paths based on real-time vocal analysis using Web Audio APIs.

An advanced retrieval-augmented generation system for large language models that learns and adapts over time. It improves retrieval accuracy by continuously learning from user interactions and feedback, making the LLM responses more precise and context-aware with every use.

A digital twin simulation of urban environments using Multi-Agent Reinforcement Learning. Multiple AI agents learn to optimize city operations such as traffic flow, energy distribution, and emergency response in a shared virtual environment powered by real-world map data.

A Flutter-based mobile platform for city bus commuters. Users get a digital bus card that works across any bus in the network. They can recharge their account through the app and enjoy seamless, cashless travel throughout the city.

A Flutter mobile app paired with an in-car diagnostic device. The device continuously monitors vehicle health metrics and streams data to the app, where a deep learning model analyzes patterns to predict failures before they happen, keeping drivers safe and reducing maintenance costs.