• 🔭 I build LLM-powered agent systems and full-stack apps — multi-agent orchestration, hybrid RAG pipelines, and end-to-end services with LangGraph / LangChain, FastAPI and Vue.js
• 🌱 M.Eng. student in Computer Science at Guangzhou University (2024–2027); my research focuses on image generation & editing with diffusion models (Stable Diffusion, CLIP, DreamBooth)
• 💬 Ask me about LLM agents, RAG optimization, FastAPI, Spring Boot, or Vue.js
• 🎯 Open to AI application / full-stack development opportunities (graduating June 2027)
• 📍 Based in Guangzhou, China

A Dify-based intelligent agent for project-report review, analyzing new proposals against historical reports and custom criteria.

• Co-built an NL2SQL agent and designed a version-based optimistic-locking mechanism to keep multi-user concurrent edits consistent
• Designed a chapter-level document chunking strategy — LLM-based heading classification with rule-based fallback for out-of-taxonomy documents
• Built a multi-stage concurrent ingestion pipeline (independent stages chained by queues) enabling cross-stage parallel processing of bulk files

A local-first AI office tool for confidentiality-sensitive archives: long-context document extraction plus CLIP-powered image search.

• Deployed the long-context Qwen2.5-1M model locally via Ollama, with Nginx multi-GPU load balancing
• Combined PaddleOCR structured recognition with dynamic prompt templates to extract targeted information via the LLM's long context window
• Designed a multi-level caching strategy (intermediate OCR states + extraction results), cutting response latency from minutes to milliseconds
• Implemented image-to-image and text-to-image search based on CLIP vision–language alignment

An AI-powered Q&A platform for pet owners: private knowledge bases, long-term user memory profiles, multi-expert collaboration, and an extensible skill-plugin marketplace.

• Multi-agent orchestration — Supervisor routing node + Specialist executors built on LangGraph, with parallel task execution and ReAct-style tool calling
• Hybrid RAG — dual-channel vector + BM25 recall fused with RRF, parent–child chunk indexing, and fine-grained reranking via qwen3-rerank
• Knowledge-graph memory — Neo4j-based long-term memory with an LLM triple-extraction pipeline, entity deduplication, community clustering, and four-layer provenance

A multi-agent travel planning system built on LangGraph, integrating the AMap (Gaode) Maps API for intelligent itinerary generation — attraction search, weather checks, and hotel filtering rolled into personalized daily plans.

• Designed a Supervisor/Specialist multi-agent architecture with shared workflow state for cross-stage information passing
• Implemented dynamic MCP tool binding — agents asynchronously discover and invoke external tools at runtime
• Exposed RESTful services via FastAPI with real-time POI, routing, and weather queries

My research focuses on image generation and editing: cross-modal representations (CLIP), diffusion models (VAE / DDPM / DDIM), personalized fine-tuning (Textual Inversion, DreamBooth), and classic editing methods (Instruct-Pix2Pix, PnP).

• 📄 EmoPrompt+: Emotional Image Content Generation via Emotion-Driven Prompting and Multi-Level Emotional Guidance in Stable Diffusion — CCBR 2025 (accepted)
• 📄 C-EmoEditor: Affective Image Editing Based on Valence-Arousal Model — The Visual Computer (under review)

Languages

LLM & Agents

Backend & Data

Tools & Infra

🔥 Keep shipping —