Ho Quang Hien

agents_rag_all_in_one. This repo outlines essential patterns and components for building modular Agentic AI systems. It also includes a full RAG pipeline and key communication protocols like MCP, A2A, and ACP.

llm_training_techs. Build transformer-based LLMs from scratch — from core components (embedding, attention, etc.) to full GPT/LLaMA models. Includes training pipelines, compression, and efficiency techniques.

A2A-samples. A multi-agent system using Google A2A and ADK to coordinate host and remote agents through MCP protocols.

text2poster. A model capable of generating product advertising banners automatically with Stable Diffusion.

sent_embed-transformer. Incorporating sentence embeddings into Transformer architecture for Neural Machine Translation

hand-gesture-recognition. A research paper on hand gesture recognition has been accepted for presentation at the 2024 9th International Conference on Intelligent Information Technology (ICIIT 2024).

mSimCSE-custom. We can create multilingual versions of SimCSE with any model using an encoder architecture by mapping cross-lingual sentences into a shared embedding space.

RAG_Agentic_Chunking_Llame3.1. A Retrieval-Augmented Generation (RAG) system built upon the Agentic Chunking Strategy