Zifei Liu

I come from wilderness and woodland.

I'm Zifei Liu, aka Zoro. Algorithm researcher at a cybersecurity company in Shanghai. Day-to-day I work on NLP, knowledge graphs, and LLM agents — essentially teaching machines to read unstructured Chinese: network traffic, threat intel, sales reports, weekly-report emails.

My tech stack has evolved from "BERT + NER" to "LLM + Agents + full-stack." I've led the solo application of a tens-of-millions-yuan provincial research project, and I'm putting large models into real production: AI video inspection, compliance review, sales analysis, weekly-report generation.

This site is a slow record of what I read, watch, and walk through — AI paper notes in Chinese, live A-share / HK-market dashboards, hiking trails, and camera moments. Algorithms are getting faster, but understanding takes time.

No algorithm. No push. No ads. If "slow" still sounds like a luxury, stay for one piece.

• ZoroTreeking 2026 -

  Personal blog + data-automation lab. Astro 4 · Multi-pipeline auto content (AI digest / A-share & HK live dashboards / ETF three-factor / photo albums / hiking GPX) · Bilingual · LLM translation + summarization pipeline.

• LLM Agent business loops (6 scenarios)

  2025 -

  From demo to production business loops. Stack based on Function Calling + multi-agent orchestration + RAG. Covers 6 real scenarios: AI video inspection (VLM anomaly detection) · Compliance review (long-doc term extraction) · Optimization algorithms (business-parameter tuning) · Crawler systems · Sales analytics · Auto weekly reports. End-to-end pipeline (data → model → eval → deploy → monitor) with significant manpower savings.

• Multi-source Threat Intel Fusion System

  2024 -

  End-to-end cybersecurity knowledge graph: local LLM fine-tuning for extraction · Multi-source fusion (+10% accuracy) · Nebula storing 1M+ entities / 10M+ relations.

• Threat Lead & Identity Mining System

  2024 -

  Mass personnel + telecom data analysis · Clustering graph algo processing 100k+ records at second-latency · Doris + Nebula real-time multi-source intel pipeline.

• Provincial Research · Threat Trail & Path Optimization

  2024

  Solo-led the full application (tens-of-millions-yuan funding): sparse computing · threat intel extraction · knowledge graph · from guideline reading to defense presentation.

• 📖
  To Live — Yu Hua

  The raw power of life itself

• 📖
  Ordinary World — Lu Yao

  How ordinary people shine

• 📖
  Those Things about the Ming Dynasty — Dangnian Mingyue

  History's inevitable and its accidents

• 📖
  Sword Snow Stride — Fenghuo Xizhuhou

  Character arcs in wuxia

• Ship an LLM agent that truly runs a business loop — not a demo, but replacing real workflows
• Evolve from «implementing algorithms» to «defining problems + shaping solutions»
• Walk 30 mountains across China · At least 4 long-distance hikes a year
• Investing philosophy: data-driven · value investing · long-hold · never stop cultivating

1. 2016
   Bachelor in CS · Hubei Police Officer Academy
2. 2020
   Master in Cybersecurity · Hubei University · NLP + Knowledge Graph focus
3. 2022
   NLP intern at Suzhou Institute · Financial UIE + sentiment analysis
4. 2023
   Joined Shanghai Feiqi · Algorithm researcher · Cybersecurity AI
5. 2024
   Led threat-intel knowledge graph (1M+ entities / 10M+ relations) · Solo application of a tens-of-millions-yuan provincial research project
6. 2025
   Shifted to LLM Agents · Production loops for AI ops / compliance / sales / weekly-report
7. 2026
   Launched ZoroTreeking — code and mountains together