Skip to content
View ganlin770's full-sized avatar

Block or report ganlin770

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ganlin770/README.md

🚀 LLM Application Engineer

2026 graduate transitioning from law to AI / Agent engineering. I specialize in shipping production-grade LLM toolchains — turning large-model capabilities into reusable tools that solve real problems. Self-built 70+ Claude Code Skills, 20+ MCP servers, and end-to-end AIGC pipelines (voice cloning, document automation, multi-platform agents).

🎯 Core Expertise

AI Engineering & LLM Integration

  • Claude / OpenAI API applications (tool use, structured output, streaming)
  • Custom MCP (Model Context Protocol) servers — design & implementation (TypeScript & Python)
  • Agent workflow orchestration & prompt engineering
  • Multi-modal AIGC: LLM + voice cloning (MiniMax, ElevenLabs, F5-TTS) + image generation

Full-Stack & Data

  • Backend: Python (FastAPI, async), Node.js / TypeScript
  • Frontend: React / Next.js, Tailwind CSS
  • Econometrics & ML: OLS / panel / DID / IV, XGBoost / LSTM (Python · R · Stata)

📌 Featured Projects

  • tool-smith — LoRA-fine-tuned a 0.5B model into a JSON tool-call router (PEFT+TRL), served over MCP, with a base-vs-tuned eval + an agent loop (validate → repair-retry → fallback). PEFT/TRL · MCP · eval
  • promptlab — from-scratch hybrid rule + LLM-as-judge eval harness: pass-rate, judge-vs-rule agreement, CI gate, JSONL audit. Python · LLM eval
  • thesis-generator — full-stack FastAPI LLM app: 8-stage pipeline (parse → analyze → LLM content → DOCX), deployed. FastAPI · Claude API
  • matlab-viz-mcp — MCP server exposing MATLAB plotting as an LLM-callable tool. Node.js · MCP SDK

🛠️ Tech Stack

Category Technologies
Languages Python, TypeScript, R, Stata, SQL
LLM / AI Claude API, MCP, Prompt Engineering, Agent orchestration, RAG
Backend FastAPI, Node.js, PostgreSQL
Frontend Next.js, React, TailwindCSS
Data / ML Econometrics, XGBoost / LSTM, matplotlib viz
DevOps Git, Docker, Zeabur / Vercel

📊 Track Record

  • 70+ Claude Code Skills — self-built automation across content, research & data workflows
  • 20+ self-built MCP servers — MATLAB viz, paper search & more (TypeScript & Python)
  • AIGC video pipeline — script → Remotion → cloned-voice TTS → mixed MP4 (bilingual)
  • Real paying users — academic data-analysis & modeling service

🎓 Background

  • Law (法学) undergrad (2026), Jiangxi University of Finance & Economics → self-taught into AI / Agent engineering & applied data analysis
  • Econometric modeling, quantitative analysis, data storytelling
  • "If it doesn't solve a real problem, it's not engineering."

📬 Get in Touch

  • 📧 Email: 2379286964@qq.com
  • 🐙 GitHub: @ganlin770

Open to LLM / AI engineering internships & roles.

Pinned Loading

  1. thesis-formatter-fullstack thesis-formatter-fullstack Public

    江西财经大学现代经济管理学院毕业论文格式化工具 - 全栈应用 (Next.js + FastAPI + Railway)

    Python 5 1

  2. matlab-viz-mcp matlab-viz-mcp Public

    MCP server that lets LLMs generate MATLAB visualizations — execute MATLAB code via Model Context Protocol, return PNG/JPG/SVG

    JavaScript

  3. paper-search-server paper-search-server Public

    MCP server for academic paper search via Elsevier Scopus API — Flask-based, caching, structured results for LLM research workflows

    Python

  4. thesis-generator thesis-generator Public

    全栈 AI 论文生成系统：FastAPI + Claude API + python-docx，8 阶段自动生成格式化 Word 论文（含 SSE 实时进度、Zeabur 部署）

    Python

  5. promptlab promptlab Public

    A from-scratch hybrid rule + LLM-as-judge eval harness: pass-rate, judge-vs-rule Cohen's kappa, latency/cost, HTML report, CI gate. Runs offline.

    Python 1

  6. tool-smith tool-smith Public

    LoRA-fine-tune a 0.5B model into a JSON tool-call router, serve it over MCP, prove the lift with a base-vs-tuned eval, wrap it in an agent loop with failure recovery. Teaching-scale, real artifacts.

    Python