I am an Electrical Engineering Ph.D. Student passionate about artificial intelligence, cognitive architectures, and agentic systems. My research and engineering focus spans the boundary of neuro-symbolic integration, reinforcement learning, and advanced multi-agent systems.

• Agentic Workflows & Multi-Agent Collaboration: Automating complex research, reasoning, and planning tasks.
• Cognitive Architectures & Neuro-Symbolic Integration: Bridging symbolic reasoning (e.g., knowledge bases, logic rules) with deep neural networks.
• Reinforcement Learning & Embodied AI: Spanning Sim2Real, robotic locomotion, and physical control constraints.
• Neuromorphic Computing & Spiking Neural Networks (SNN): Exploring energy-efficient computation for predictive maintenance and edge devices.

An AI-driven multi-agent research assistant designed to automate hypothesis generation, data analysis, visualization, and report writing.

• Key Architecture: Implements a modular agent framework (using LangChain, LangGraph, and OpenAI GPT) coordinating specialized agents for code execution, data visualization, and academic writing.
• Applications: Streamlines end-to-end scientific research pipelines, reducing complex manual analysis down to automated iterations.

A suite of Claude Code skill tools covering the entire academic research lifecycle.

• Features:
  • Lab Skills: Structured logs management (research-log), pptx report slide generator (report-slides), and session routing (research-mode).
  • Academic Research: Multi-agent literature review agent (deep-research), automated paper writer (academic-paper), peer review simulator (academic-paper-reviewer), and end-to-end research orchestrator (academic-pipeline).

A stigmergy-based communication protocol for decentralized multi-agent coordination.

• Mechanism: Allows independent AI agents to coordinate, share insights, and signal risks by depositing "invisible pheromones" (metadata/signals) directly on files without modifying their primary content.
• Concept: Mimics biological stigmergy to reduce centralized coordination bottlenecks in collaborative agent teams.

An enterprise-grade, high-performance private cloud file management system.

• Tech Stack: FastAPI (Backend) + React / Vite (Frontend).
• Features: Drag-and-drop tree navigation, TUS-based chunked resumable uploads for GB-level files, and seamless Cloudflare Tunnel integration for secure router-less edge deployment.

An automated homework formatter and LLM-based grading pre-processor.

• Functionality: Automatically converts student homework formats (Jupyter Notebooks, Word, PowerPoint) into clean PDFs and Base64 images to maximize rendering accuracy for downstream LLM grading pipelines. Includes batch-evaluation pipelines.

A real-time, WebSocket-based multiplayer verbal battle game featuring an LLM referee.

• Architecture:
  • Backend: FastAPI for game state and WebSockets.
  • Frontend: React / Vite.
  • Core Inference: A local vLLM running a multimodal Gemma model acting as a "toxic referee" to compute damage and return witty critiques. The core inference module is compiled with Cython for maximum speed.

A feature-rich multi-modal Discord assistant.

• Capability: Integrates large language models for voice interaction, image generation (text-to-image), visual question answering (VQA), conversation history memory, music playback, and channel RAG.

Developing personalized recommendation systems to enhance user experience.

• Email: james911129@gmail.com
• GitHub Profile: github.com/starpig1129