Appau Gideon Kofi Amo

About

I am a Master's student at the University of Science and Technology of China (USTC), focusing on intelligent systems that combine retrieval, reasoning, and adaptive memory. My current research centers on personalized agentic RAG systems with autonomous decision-making and self-correction loops.

I am also deeply interested in reinforcement learning for LLM agents, long-tailed recognition, and sharpness-aware minimization (SAM) methods for imbalanced datasets. I enjoy building end-to-end systems and open-sourcing my work when possible.

Research Vision & PhD Goals

My Master's work on agentic RAG and multi-agent orchestration has convinced me that the next frontier is not just better models, but better agents — systems that can perceive, reason, act, and learn in open-ended environments. A PhD will give me the depth and mentorship to pursue this at the highest level.

Specific Research Directions

• RL for LLM Reasoning: Train language models to reason through verifiable rewards (GRPO, PPO) rather than supervised fine-tuning alone — enabling self-improving agents for math, code, and planning.
• Continual & Lifelong RL: Design agents that learn new tasks sequentially without forgetting prior knowledge, using techniques like EWC, progressive networks, and episodic memory.
• Embodied AI & Robotics: Bridge high-level LLM reasoning with low-level robotic control through Vision-Language-Action (VLA) models and skill discovery.
• Multi-Agent Coordination: Extend my Agent Prime work to decentralized multi-agent systems with economic incentives, reputation, and emergent collaboration.

I am actively looking for PhD advisors working in these areas, particularly at institutions with strong RL and robotics groups. I am open to positions in mainland China, Hong Kong, and internationally.

Research Interests

Selected Projects

• Personalized Agentic RAG System
  2025 - Present

  Designing memory-augmented agentic RAG with adaptive retrieval, user modeling, confidence estimation, and self-correction loops. Built with LangGraph, LangChain, FAISS, and LLMs.

  Code
• Campus Knowledge Base RAG Chatbot
  2025

  A RAG prototype for campus knowledge retrieval using Ollama embeddings, Chroma vector database, and a Gradio-based user interface.

  Code
• Class-Adaptive Focal SAM
  Master's Thesis, 2025–2026

  A class-adaptive focal loss combined with Sharpness-Aware Minimization (SAM) for long-tailed visual recognition. Evaluated on CIFAR-10/100-LT, TinyImageNet, ImageNet-LT, and iNaturalist 2018.

  Thesis — Code coming soon
• Agent Prime - Multi-Agent Orchestration with USDC Nanopayments
  Lablab.ai Agentic Economy on Arc Hackathon, 2026

  A 3-agent workflow (Research / Writer / Editor) with real USDC nanopayments on Arc Testnet. Includes on-chain verification via web3.py and hits both Usage-Based Compute Billing and Agent-to-Agent Payment Loop tracks.

  Code
• Resume / Candidate Screening System
  Future Interns - Remote Internship, Mar–Apr 2026

  An ML-powered resume screening and candidate ranking system built during a remote internship. Uses NLP and transformer-based models for automated candidate evaluation.

  Code
• RL Portfolio - Dueling DQN, PPO, A2C, VPG
  Reinforcement Learning / 2026

  Hands-on implementation of Dueling DQN and modern RL algorithms based on ARENA, UC Berkeley CS 285, and Spinning Up. Exploring RL for LLM agents and mechanistic interpretability.

  Code
• Weather Trend Forecasting
  Time Series / ML / 2025

  Weather trend forecasting using time series analysis and ML models.

  Code
• AI Study Coach
  EdTech / NLP / 2024

  An interactive AI-powered study assistant with personalized learning paths, quiz generation, and progress tracking using LLM-based techniques.

  Code

Blog

• From Tutorial to Production: Building a Modular RLHF Pipeline
  July 2026

  A deep dive into transforming three tutorial scripts (SFT → Reward Model → PPO) into a production-grade, configurable RLHF framework. Covers Bradley-Terry loss, clipped PPO objectives, KL divergence penalties, and lessons learned from training a 0.5B model end-to-end.

  Read Post Code

Publications

Coming soon — publications in preparation.

Experience

• Mar – Apr 2026

  Machine Learning Intern

  Future Interns (Remote)

  Built a Resume/Candidate Screening System using NLP and transformer models.
• Nov 2022 – Aug 2023

  IT Assistant

  AAMUSTED - Akenten Appiah-Menka University of Skills Training and Entrepreneurial Development

  Provided IT support, system maintenance, and technical assistance.

Education

• 2024 – 2027 (Expected)

  M.S. in Software Engineering (AI/ML)

  University of Science and Technology of China (USTC)
• 2018 – 2022

  B.Sc. in Information Technology Education

  University of Education, Winneba

  First Class Honors

Technical Skills

Certifications

• NVIDIA - LLM RAG Agent Fundamentals (2025)
• DeepLearning.AI - Advanced Learning Algorithms (2025)
• DeepLearning.AI - Supervised Machine Learning (2025)

News