- Long-Tailed Recognition — Class-Adaptive Focal Loss + Sharpness-Aware Minimization (SAM)
- Reinforcement Learning — Dueling DQN, PPO, A2C, VPG, RLHF
- Embodied AI & Robotics — Multimodal agents, robot learning, world models
- LLM Agents & RAG — Agentic systems with real on-chain payments (USDC)
| Project | Description | Tech |
|---|---|---|
| rl-portfolio | RL algorithms from scratch: Dueling DQN, PPO, A2C, VPG in PyTorch | PyTorch, Gymnasium |
| agentic-nano-ai | Agentic AI experiments: multi-agent systems and autonomous task agents | Python, FastAPI |
| Campus-Knowledge-Base-RAG | RAG chatbot for campus knowledge base — retrieval-augmented generation | Python, LLM, RAG |
| AI-Study-Coach | AI-powered interactive study assistant with personalized learning paths | Python, NLP |
| FUTURE_ML_02 | Resume/Candidate Screening System — automated resume parsing & ranking | Python, ML, NLP |
| weather-trend-forecasting | Weather trend forecasting using time series analysis and ML models | Python, Time Series |
Languages: Python, LaTeX, SQL, JavaScript
ML/DL: PyTorch, Transformers (Hugging Face), FAISS, wandb, SwanLab
RL: Gymnasium, Stable-Baselines3
Tools: FastAPI, Streamlit, Jupyter, VS Code, Git, Linux
Cloud: AutoDL, Google Cloud, AWS (learning)
- 🧠 Embodied AI — JEPA, world models, robot learning (ManiSkill, Isaac Sim)
- 🎨 Diffusion & Flow Models — For image/video generation
- 🗣️ Multimodal AI — Tokenizer-free foundation models, VLM from scratch
- 🇨🇳 HSK3 Chinese — Preparing for August exam
- 🌐 Portfolio: amo-gideon.github.io
- 💼 LinkedIn: add your link
- 🐦 X/Twitter: add your handle if you have one
> "Building from first principles — one paper implementation at a time."