catkhanh

Developed Reinforcement Learning and Machine Learning models focusing on dynamic fee adjustments to improve trading efficiency and swap rates in decentralized finance (DeFi) platforms. Designed and conducted extensive backtesting frameworks to evaluate the performance of liquidity provision strategies under varying market conditions. Built a model for liquidity providers (LPs) that optimizes capital efficiency by dynamically adjusting liquidity positions based on real-time market data and fee structures. Simulation results demonstrated the agent’s effectiveness in balancing volume capture and fee optimization, achieving near-optimal trade-offs.

Conducted research and developed Machine Learning projects in the finance domain to provide intelligent trading suggestions for ENSA Chatbot users within the Entrade X app. Deployed systems for real-time daily stock market analysis utilizing technical indicators to provide actionable insights and support investment decisions. Worked with Apache Kafka and Faust Streaming to build scalable, real-time financial data pipelines that deliver high-performance, low-latency solutions.

Teaching Assistant of the course Programming Techniques in C/C++

Build alphas utilizing WorldQuant’s BRAIN API, leveraging its extensive financial data, simulation tools, and advanced research features. Program an automated research pipeline to systematically discover, test, and validate new alphas, improving efficiency in signal discovery and evaluation. Utilize WorldQuant’s BRAIN platform, a specialized environment for alpha research, which provides vast financial datasets, backtesting frameworks, and sophisticated tools for consultants to refine trading strategies.

Worked on projects using OCR to read data from medical prescription for the app OurHealth

Focused on research and development of robotic grasp detection systems, emphasizing real-time performance and multi-object grasping capabilities. Developed the Grasp-YOLOv8 model, a one-stage robotic grasp detection framework that achieved state-of-the-art accuracy on the Cornell dataset (98.9%) and the CIM900 dataset (91.9%), with real-time processing speed (23.8 FPS).