Projects

Innovative AI and Full-Stack Development Work

10 Total Projects

Detailed showcase of my development and research work

AI Doctor Assistant

Featured

A comprehensive SaaS platform that revolutionizes medical consultations through AI-powered voice interactions, providing intelligent symptom analysis and automated report generation.

Key Features:

  • LLM-based symptom analysis for natural doctor-patient conversations
  • Real-time speech understanding with VAPI Voice Assistant + OpenRouter AI
  • Automatic medical report generation (symptoms, prescriptions, advice, summary)
  • Clerk authentication for secure access
  • NEON PostgreSQL + Drizzle ORM for robust data operations
  • SaaS-ready architecture: dashboards, doctor listings, pricing, health tips
  • Deployed on Vercel for global accessibility

Technologies:

Next.jsTypeScriptAI/LLMPostgreSQLVoice AIVAPIClerk AuthDrizzle ORMSaaS

TimeClipAI: Real-Time Action Classification

Featured

Advanced deep learning framework for real-time video analysis, achieving state-of-the-art performance in temporal action localization across multiple benchmark datasets.

Key Features:

  • Real-time action classification and time segmentation in videos
  • Support for EGTEA, EPIC-Kitchen 100, THUMOS'14, CricShot10 datasets
  • Pre-trained I3D features for efficient training/testing
  • Offset Scoring Network (OSN) for refined action boundary predictions
  • State-of-the-art performance with robust generalization
  • Optimized for both accuracy and inference speed

Technologies:

PyTorchPythonComputer VisionTransformersVideo AnalysisDeep LearningI3D Features

DeshiPlate AI: Bangladeshi Food Recognition

Featured

An intelligent food recognition platform leveraging deep learning to identify 33 traditional Bangladeshi dishes, providing instant nutritional insights and personalized health recommendations.

Key Features:

  • AI-powered recognition of 33 traditional Bangladeshi dishes
  • Custom-curated dataset of Bangladeshi cuisine for model training
  • NextViT deep learning architecture achieving 89.76% accuracy
  • Real-time inference engine for instant food classification from images
  • Personalized nutrition recommendations based on classified dishes
  • Health suggestions tailored to identified food items
  • NEON PostgreSQL + Drizzle ORM for scalable data operations
  • Modern, responsive UI deployed on Vercel

Technologies:

Next.jsTypeScriptNextViTPyTorchDeep LearningPostgreSQLDrizzle ORMOpenRouter AIComputer Vision

SHOBDOTORI: Regional-to-Standard Bangla Speech Recognition

Featured

Developed an Automatic Speech Recognition (ASR) platform that transcribes diverse regional Bangladeshi dialects into standard Bangla. Leveraging transformer-based speech models, phoneme alignment, audio augmentation, and n-gram post-processing, it achieves high transcription accuracy and competitive performance on benchmark evaluations.

Key Features:

  • Automatic transcription of 20 regional Bangladeshi dialects into standard Bangla
  • Curated dataset with 3,800+ audio samples
  • Transformer-based models: Whisper and Wav2Vec2
  • Phoneme alignment and audio data augmentation for robust learning
  • N-gram KenLM post-processing for fluent and linguistically correct transcriptions
  • Performance evaluation using Normalized Levenshtein Similarity (NLS)
  • AI Hackathon project under Televerse 1.0, CUET (Department of ETE)

Technologies:

PythonPyTorchWhisperWav2Vec2KenLMASRDeep LearningAudio Processing

InterviewBD: Interview Preparation Platform

Featured

A comprehensive interview preparation ecosystem where professionals share authentic interview experiences, enabling job seekers to prepare effectively with field-specific questions, verified contributors, and collaborative learning features.

Key Features:

  • Hierarchical content organization with fields, categories, and Q&A system
  • Profile verification and badge system for credible contributors
  • Real-time messaging with online status and activity tracking
  • Three-tier role management: Admin, Moderator, and User permissions
  • Advanced search, filtering, and like/bookmark functionality
  • Comprehensive moderation tools with reporting and analytics dashboard
  • JWT authentication with email verification and PostgreSQL + Drizzle ORM
  • Deployed on Vercel with fully responsive design

Technologies:

Next.jsTypeScriptPostgreSQLDrizzle ORMJWT AuthReal-time ChatNodemailerTailwind CSSVercel

Daily English Vocabulary Platform

Featured

A comprehensive English learning platform that leverages Google Gemini AI to generate daily vocabulary tests, featuring intelligent progress tracking, sequential learning paths, and personalized word-saving capabilities for effective language mastery.

Key Features:

  • AI-generated daily vocabulary tests using Google Gemini 3.0 Flash
  • Registration date-based test access with sequential learning enforcement
  • Comprehensive progress tracking with score history and analytics
  • Save word feature with full CRUD operations and search functionality
  • Personal notes system for saved words with case-insensitive filtering
  • JWT-based authentication with secure password hashing
  • Automated daily test generation via Vercel Cron Jobs
  • Neon PostgreSQL + Drizzle ORM for persistent data storage

Technologies:

Next.jsTypeScriptGoogle Gemini AIPostgreSQLDrizzle ORMJWT AuthTailwind CSSVercel Cron

Educational Chatbot using RAG

Featured

An innovative educational tool leveraging Small Language Models and RAG architecture to provide accessible learning support for underserved communities, with offline capability.

Key Features:

  • Lightweight SLM architecture for device-efficient operation
  • Web-based interface for easy accessibility
  • 75% accuracy on 4,000-question dataset validated with RAGChecker
  • Offline capability for areas with limited internet connectivity
  • Supports underserved communities in learning
  • Retrieval-Augmented Generation for accurate responses

Technologies:

PythonRAGSLMLangChainOllamaFlaskNLP

Road Accident Analysis System

Comprehensive machine learning pipeline for emergency response optimization, comparing multiple algorithms to achieve high accuracy in critical medical predictions.

Key Features:

  • Predictive models for patient status and injury type classification
  • Comparative analysis: Logistic Regression, Decision Tree, SVM, XGBoost, Random Forest
  • Achieved 99.1% accuracy for patient status prediction
  • 89% accuracy for injury type classification
  • Data preprocessing and feature engineering pipeline
  • Insights for improving emergency response strategies

Technologies:

PythonScikit-learnXGBoostPandasNumPyMLData Analysis

Agile-Driven Movie Reservation Website

Professional web application demonstrating modern development practices, built by a 5-member team following Agile principles with comprehensive testing and documentation.

Key Features:

  • Complete movie browsing and reservation system
  • Frontend/backend architecture with MVC pattern
  • RESTful API design for seamless communication
  • Comprehensive code documentation using JSDoc
  • Unit and integration testing with Jest
  • Agile Scrum methodology with sprint planning and retrospectives
  • Team collaboration using Git and Trello

Technologies:

MongoDBExpress.jsReact.jsNode.jsMVCJestJSDocAgile

Viper Travel Booking Website

Complete travel management system with robust database operations, providing users with seamless booking experiences and travel planning capabilities.

Key Features:

  • Comprehensive travel booking functionality
  • SQL-based CRUD operations for data management
  • User authentication and session management
  • Search and filter capabilities for destinations
  • Booking history and management features
  • Responsive design for mobile and desktop

Technologies:

PHPMySQLJavaScriptHTML/CSSSQLCRUD API