Anh Nguyen

Hi, I'm Anh Thach Ha Nguyen (Apricity) πŸ‘‹

|

I’m currently pursuing a Master’s degree in AI, focusing on LLMs and deep learning. My goal is to apply these techniques to extract insights from financial data and build robust signal-driven models. I’m a curious researcher who thrives on solving complex problems and learning from experimentation.

πŸ“ View Projects πŸ“„ View CV (PDF) 🌐 GitHub Profile

πŸ‘€ About Me

2025 – Present

πŸŽ“ Pursuing Master’s degree in Artificial Intelligence at University of Science (HCMUS). Focus on LLMs, Time Series Forecasting, and Reinforcement Learning for Finance.

2021 – 2025

πŸ“˜ Studied Artificial Intelligence (B.Sc.) at FPT University, Ho Chi Minh City. Gained solid foundations in machine learning, deep learning, and algorithmic thinking, with hands-on experience in AI applications.

My Research Interests

🧠 LLMs, Time Series Modeling, RLHF, Quantitative Finance, Signal Modeling, Autonomous Agents for Finance.

Career Goal

πŸš€ I aim to develop AI models that understand and optimize global financial strategies through research, experimentation, and impact-driven innovation.

πŸ› οΈ Skills

AI / ML / LLM

Data & Dev

Programming

Maths / Statistics

Quant Finance

πŸ“Œ Featured Projects

πŸ“Š Financial Text Signal Generator

RL Agent

Fine-tuned GPT-2 to generate market sentiment signals from financial news headlines.

  • Stack: Python, HuggingFace, yFinance, GPT-2
  • Best Result: ROUGE-L = 0.68, BLEU = 0.42

πŸ€– RAG QA for Earnings Calls

RL Agent

Built a LangChain-based agent to answer earnings call questions using hybrid search (FAISS + OpenAI).

  • Stack: LangChain, OpenAI API, FAISS, Streamlit
  • Demo: Answer accuracy ~89% on 20 test queries

πŸ“ˆ RL for Portfolio Optimization

RL Agent

Trained PPO and DDPG agents to allocate assets in synthetic markets with Gym environments.

  • Stack: Stable-Baselines3, PyTorch, Gym, NumPy
  • Reward: Sharpe Ratio ↑ 12% vs baseline strategy

πŸ“š Publications

The Impact of Varying Knowledge on Question-Answering System

Authors: Anh Thach Ha Nguyen, Trung Nguyen, et al.

Published at: 2024 Asian Conference on Communication and Networks (ASIANComNet), IEEE

Summary: Proposed a knowledge-based question answering system using DPR and Fusion-in-Decoder (FiD) architecture to enhance response quality and reduce hallucination. Evaluated on ELI5, MS MARCO, and MASH-QA datasets.

πŸ“„ View Full Paper (PDF)
DOI: 10.1109/ASIANComNet63184.2024.10811070

πŸ“¬ Contact Me

Email: anhthachhanguyen.contact@gmail.com

LinkedIn: /helloapricity

GitHub: /helloapricity

"I believe in building intelligent systems that learn from real-world data and generate measurable impact."