Ahmed Khamis
Researcher · MSc at Georgia Tech
I work on large language models and machine learning research. My focus is post-training: reinforcement learning, alignment, and weight space programming. These are my notes on what I build and test.
Writing
2025
- LLM Agentic Memory Systems Why agents forget, and four kinds of memory that help.
- LLM Alignment: SFT, RL, and ORPO Classic alignment against Odds Ratio Preference Optimization.
- CUDA Lab Series 2 - Getting Started with CUDA: A Practical Guide Threads, blocks and grids, and how CUDA differs from C.
- CUDA Lab Series 1 - Parallel Computing and CUDA Sequential against parallel, CPU against GPU.
2024
- Research Digger: Streamlining Academic Research with AI Fetching, analysing and summarising papers with chain-of-thought prompts.
- Pooling Encoder-only Transformers Embeddings for Downstream Tasks Turning encoder embeddings into strong classifiers and regressors.
- Fine-tuning Llama3 Models with LoRA on Custom Data Low-rank adaptation, then a full run on Llama 3 8B.