Duration: Apr 2024 - Jun 2024

Description: Developed an intelligent image resizing tool using Java, implementing the Seam-Carving algorithm with a user-friendly GUI. Users could mark protected areas and prioritize regions for content preservation during resizing.

Key Achievements:

• Implemented the core algorithm using dynamic programming for efficient seam identification and removal
• Designed and developed the GUI for interactive image operations
• Optimized algorithm performance for large-scale image processing

Outcome: Achieved an excellent score (100/100)

Duration: Jun 2024 - Jul 2024

Description: Built a website for stock data query, simulated investment, and price prediction, leveraging TradingView APIs for real-time data visualization and analysis.

Key Achievements:

• Integrated TradingView APIs for dynamic stock data display
• Developed a simulated investment module with a backtracking strategy for performance evaluation
• Embedded price prediction visualization and personalized stock recommendations

Outcome: Received an individual grade of A-

Duration: Sep 2024 - Dec 2024

Description: Designed a Google Scholar-like academic publication management system, managing data import, performance optimization, and permission management.

Key Achievements:

• Implemented data import from 2.7GB .ndjson files and optimized import efficiency
• Conducted performance analysis by comparing database operations and file I/O
• Integrated role-based permission control (e.g., admin, user) and login authentication
• Developed a frontend interface using JavaScript and Java APIs

Outcome: Scored 100/100 + 10 (bonus) in this project

Duration: Apr 2025 - Jun 2025

Description: Led the fine-tuning of the Qwen2.5-0.5B-Instruct model to specialize in complex Text-to-SQL conversion, emphasizing efficient memory management and advanced model optimization techniques for deployment readiness.

Key Achievements:

• Data Strategy: Enhanced model generalization through research and implementation of dataset diversification techniques
• Memory Optimization: Applied LoRA (Low-Rank Adaptation) to resolve excessive GPU memory usage, mitigating a critical training bottleneck
• Distributed Training: Integrated Ray for distributed computing, ensuring training stability and enabling efficient large-scale deployment

Outcome: Scored 100/100 in this project