AGENTS
MiniBot
A small MiMo-powered agent with tool calls, context compression, and file-backed memory. It is a practical place to test how an agent's control loop is organized.
View repository ↗Projects
From research prototypes to practical systems.
AI ENGINEERING
Enterprise RAG Assistant
A local knowledge-base assistant built with Next.js and FastAPI. It combines dense and keyword retrieval, lightweight reranking, and source-linked answers.
View repository ↗SYSTEMS
CUDA GEMM
A custom PyTorch CUDA extension for matrix multiplication. The repository documents tiling, shared-memory reuse, register tiling, and profiler-based comparison.
View repository ↗MULTIMODAL
ViT–GPT-2 Video Captioning
My master's project: a ViT-based visual encoder and GPT-2 decoder for video captioning, with local inference, benchmarking, and profiling tools.
View repository ↗