Building things with LLMs, one commit at a time.
I love to talk on trending ⚡ AI / LLM technology ⚡, stay hungry stay foolish.
Intelligent E-Commerce Customer Service & RAG Policy QA System
- Tiered Model Architecture: lightweight tasks (intent/slot/rerank/summary) on qwen3.7-flash (200
300ms), complex reasoning on qwen3.8-max; disabled deep thinking to cut latency from 69.8s to 310s - Three-layer Memory Engine: Write-Through to MySQL + Redis, 12-turn incremental summary (≤250 chars), ~35% token reduction
- Anti-hallucination RAG: query rewrite → dual-mode embedding (DashScope / Feature Hashing fallback) → rerank Top-4 → forced [1][2] citation, hallucination rate 30% → <5%
- Stack: Python · FastAPI · LangChain · MySQL · Redis · Chroma · Streamlit · DashScope
Automated screen recording & OCR-naming asset manager for mobile games, supports KernelSU / SukiSU / Magisk.
AI / LLM
Backend
DevOps & Tools
Stay hungry, stay foolish. Always learning, always building.