• LLM Infrastructure: Deployment and optimization using vLLM and FastGPT.
• Autonomous Agents: Designing multi-skill agents based on the MoltBot framework.
• Edge CV: Real-time vision tasks on NVIDIA Jetson and Rockchip platforms.
• System Integration: Bridging the gap between digital intelligence and physical control.

• LLM Deployment: vLLM, Qwen, DeepSeek, model quantization.
• RAG & Knowledge: FastGPT, vector database integration.
• Agent Frameworks: Tinbot development, Browser Automation, Task Planning.

• Models: YOLOv10, real-time object detection.
• Acceleration: TensorRT C++ deployment, model distillation.

• Edge Platforms: NVIDIA Jetson Orin NX, RK3588S.
• Embedded: ESP32-S3, UART/I2C/SPI interface protocols.
• DevOps: Docker, Linux system optimization, CI/CD pipelines.

End-to-End Robotics Software Architecture · Core Developer

A fully integrated physical robot seamlessly combining voice interaction, vision, and mechanical control. (Closed-source company product)

• Intelligent Workflow: Engineered a closed-loop system: Voice Wake-up -> ASR -> Agent Dispatch -> Robot Execution -> TTS Broadcast.
• Vision & Grasping: Utilized Multi-modal LLMs, FastSAM + TensorRT for real-time segmentation, and depth camera inverse kinematics for precise object grasping.
• Cognitive & Navigation: Integrated FastGPT for online knowledge base dialogues and empowered the robot with autonomous navigation tasks.

Autonomous Multi-Skill AI Agent · Lead Developer

Bridging browser automation and local OS control.

• Built on MoltBot architecture to handle high-complexity, multi-step tasks.
• Integrated native browser automation for dynamic data retrieval.

High-Performance Local RAG Deployment Solution · Creator

A one-stop deployment architecture solving VRAM OOM and network constraints.

• Inference & Search: Built with vLLM (Qwen2.5-7B) and TEI (BGE-M3) to achieve high-concurrency and low-latency queries.
• VRAM Optimization: Fine-tuned model parameters to stably co-host LLM and Embedding models on a single GPU.
• Network & Access: Customized for domestic network environments (HF/Docker acceleration) and integrated FRP + Nginx for secure public access to intranet services.

• AI R&D Engineer in a Shenzhen-based startup.
• Focused on full-stack AI implementation from data annotation to deployment.

• Embodied AI: Bringing Large Multi-modal Models (LMM) into physical hardware.
• System Efficiency: Maximizing AI performance on resource-constrained edge devices.
• Developer Experience: Building reusable AI components and automated workflows.