OmniPatrol-VLA: 基于 Qwen2-VL 与 ROS2 的分层具身决策巡检系统。支持 LoRA 微调与 4-bit 量化，实现端云协同的智慧交通违章自动研判。 (English: A multi-modal smart traffic inspection robot powered by Qwen2-VL and ROS2. Features LoRA fine-tuning, 4-bit quantization, and edge-cloud hierarchical architecture.)

🚀 Advanced multi-agent autonomous framework for deep research, self-reflective peer review, and citation-accurate report compilation. Powered by LangGraph state machines, ChromaDB local memory, and fine-tuned Unsloth LoRA models. 🔍📚🧠

Spam Email Classification using LoRA Fine-tuned Transformers: High-performance spam email classification using LoRA-adapted transformer models (ELECTRA, RoBERTa). Achieves 99.4%+ accuracy with parameter-efficient fine-tuning on 83K+ emails.

An open-source, text-guided image generation system for agricultural crops based on Stable Diffusion v1.5 with LoRA fine-tuning and a Streamlit UI.

DecompGTI is a graph-reasoning framework that fine-tunes small language models to parse natural-language graph queries into structured JSON plans executed by deterministic algorithms on a FastMCP server.

Production-grade multi-model LLM inference + LoRA fine-tuning on bare-metal Slurm clusters (8× B200). SGLang, vLLM, EAGLE, MTP speculative decoding.