Developed an AI-powered IoT smart mirror with Raspberry Pi 4, featuring intelligent makeup simulation using GAN models, voice control with Google Assistant API, and comprehensive information display including weather, stocks, and calendar integration.

AI & Computer Vision Features

• BeautyGAN Integration: Real-time makeup transfer using pre-trained GAN model for virtual cosmetics experience
• Face Detection & Processing: Camera-based user recognition and image processing pipeline
• Voice AI Control: Google Assistant API integration for hands-free interaction
• Motion Detection: UV4L and motion API for smart activation and security monitoring

IoT Hardware & Software Integration

• Raspberry Pi 4 System: Complete embedded system with camera, display, and sensor integration
• Magic Mirror Framework: Node.js/Electron-based application with custom UI modules
• Real-time Information Display: Weather forecasts, stock prices, calendar events, and email notifications
• Web Streaming CCTV: Live video streaming capability with motion detection alerts

Technical Achievement

Awarded 3rd place in Capstone Design Competition at Yonsei University. Successfully integrated cutting-edge AI technologies with practical IoT hardware to create a functional smart home device that enhances daily routines through intelligent automation.

Developed an intelligent multilingual chatbot for cryptocurrency Q&A using advanced NLP models including BERT, TF-IDF, Word2Vec, and SIF embeddings, with real-time price integration and automatic language translation.

AI/NLP Innovation

• Multi-Model Performance Analysis: Compared 4 embedding approaches with BERT achieving 85%+ cosine similarity accuracy
• Sentence Transformers Integration: Implemented state-of-the-art 'distilbert-base-nli-max-tokens' for semantic understanding
• Smart Answer Retrieval: Cosine similarity-based matching from 16,836 Quora Q&A pairs with quality thresholding
• Real-time Data Integration: Dynamic cryptocurrency price fetching via web scraping

Multilingual & User Experience

• Automatic Language Detection: Google Translate API integration supporting multiple languages (English, Japanese, Chinese)
• Seamless Translation Pipeline: Question translation to English, processing, and answer translation back to user's language
• Interactive Console Interface: User-friendly command-line chatbot with continuous conversation flow

Technical Excellence

Demonstrates proficiency in advanced NLP techniques, multilingual processing, and practical AI application development. The project showcases ability to handle large datasets, implement multiple ML models, and create production-ready conversational AI systems.

placeholder

Key Features

• placeholder

Technical Implementation

placeholder

Developed PMNet, a neural network specifically designed for large-scale pathloss prediction that won 1st place in the IEEE ICASSP 2023 First Pathloss Radio Map Prediction Challenge.

Key Innovations

• Custom neural architecture for pathloss prediction
• Advanced data augmentation and fine-tuning techniques
• Network architecture optimization for superior performance
• RMSE of 0.02569 on RadioMap3Dseer dataset

Competition Success

Achieved 1st place in the prestigious IEEE ICASSP competition, outperforming all competitors with the highest accuracy in radio map prediction challenges.

Developed novel Vector Quantization-based Variational Autoencoder approaches for robust massive MIMO channel prediction, achieving significant performance improvements in noisy communication environments.

Research Contributions

• Novel VQ-VAE architecture for robust communication systems
• 10% improvement in noise-robustness with 50% compression
• Up to 15 dB NMSE gains over standard autoencoders
• Practical deployment framework for 5G/6G systems

Recognition

Published in IEEE Transactions on Wireless Communications and contributed to 1st place victory in IEEE ICASSP Signal Processing Grand Challenges, demonstrating the practical impact of the research.

Developed a sophisticated multimedia system that automatically extracts hierarchical structure from video content and provides interactive navigation through an intelligent media player, combining video processing, image analysis, and professional GUI development.

Advanced Video & Image Processing

• Automatic Shot Detection: Mean Absolute Difference (MAD) algorithm for detecting camera cuts and transitions
• Scene Segmentation: Multi-modal approach using color histograms and HOG features with cosine similarity for semantic grouping
• Audio-based Subshot Analysis: Short Time Fourier Transform (STFT) of audio signals to detect mood and emotional changes
• Hierarchical Content Organization: Three-level structure (Scene → Shot → Subshot) with temporal indexing

Interactive Multimedia Player

• Professional GUI: PyQt6-based interface with hierarchical content navigation and real-time video display
• Synchronized Playback: Multi-threaded architecture ensuring perfect audio-video synchronization
• Interactive Navigation: Click-to-jump functionality with real-time segment highlighting and playback control
• Raw Data Processing: Direct RGB/WAV file handling with optimized binary operations

Technical Excellence

Demonstrates mastery of multimedia systems engineering, combining video/image processing algorithms, real-time systems programming, and professional GUI development. Successfully handles the complete pipeline from low-level media analysis to high-level user interaction.

Developed WikiSER, a high-quality dataset of 1.7M sentences with 79K labeled software entities, and implemented noise-robust learning frameworks for improved NER performance.

Dataset Contributions

• Comprehensive lexicon with 79K unique software entities
• 1.7M high-quality labeled sentences
• 12 fine-grained software entity categories
• Significantly lower error rate than existing datasets

Technical Innovation

Implemented self-regularization noise-robust learning approach achieving 7.1% F1 score improvement over state-of-the-art models and 8.62% error reduction compared to existing benchmarks.

placeholder

System Features

• RESTful API backend with Node.js and Express.js
• Modern React frontend with TypeScript

Placeholder

placeholder

Developed a native Android application for business search and discovery, featuring enhanced mobile-specific functionality including reservation management, native map integration, and touch-optimized user interactions.

Mobile-First Features

• Reservation System: Complete reservation management with create, view, edit, and swipe-to-delete functionality
• Native Map Integration: Google Maps API integration with location services and business mapping
• Fragment-Based Architecture: Modular UI with MapFragment, ReviewFragment, and BusinessDetailFragment
• Touch-Optimized UX: Gesture-based interactions including SwipeToDeleteCallback for intuitive mobile experience

Android Development Excellence

• Material Design Implementation: Modern Android UI following Material Design principles
• RecyclerView Optimization: Efficient list management with custom adapters for businesses, reviews, and reservations
• Activity + Fragment Pattern: Scalable architecture with BusinessDetailActivity and modular fragments
• Auto-Suggestion System: Real-time search suggestions with AutoSuggestAdapter

Technical Implementation

Demonstrates native Android development expertise with Java, implementing complex UI interactions, location services, and data management. Features splash screen, tabbed interfaces with ViewPager, and responsive design optimized for mobile form factors.

Comprehensive networking systems project implementing a complete network stack from router to transport layer, culminating in a custom TCP variant with Google's BBR congestion control algorithm.

Core Network Technologies

• IP Router Implementation: Built packet forwarding engine with ARP resolution and ICMP messaging
• Routing Protocol Configuration: Implemented BGP/OSPF for inter-domain and intra-domain routing
• Custom TCP Protocol: Developed reliable transport with stop-and-wait and sliding window mechanisms
• Advanced Congestion Control: Integrated Google's BBR algorithm for bandwidth and RTT optimization

Technical Achievements

Successfully created a full networking stack capable of interoperating with standard web servers, demonstrating expertise in network protocols from Layer 2 through Layer 4 with modern performance optimization techniques.

Developed a multi-server meeting scheduling system using C socket programming, implementing a 4-phase architecture with TCP/UDP protocols to coordinate availability across multiple backend servers and manage complex time intersection calculations.

System Architecture

• 4-Phase Communication Protocol: Boot-up → Forward → Schedule → Reply phases with precise timing coordination
• Multi-Process Coordination: Main server (TCP) orchestrates two backend servers (UDP) for distributed processing
• Hybrid Protocol Design: TCP for client-server reliability, UDP for optimized inter-server communication
• Dynamic Port Management: Automatic port assignment and discovery for flexible deployment

Advanced Algorithms

• Time Intersection Engine: Efficient algorithm to compute overlapping availability across multiple users
• Meeting Registration System: Validates proposed meeting times against availability constraints
• User Existence Validation: Pre-processing to handle non-existent users gracefully
• Concurrent Request Handling: Multi-client support with proper resource management

Technical Excellence

Demonstrates mastery of systems programming in C, including socket APIs, process management, memory handling, and protocol design. Features robust error handling, timeout management, and clean resource cleanup, showcasing production-ready networking code architecture.