Focus: Chinese NLP (Spelling Correction), Over-correction Control, Reliable LLM Inference.
I specialize in translating research concepts into reproducible, evaluable, and deployment-ready systems.
🇨🇳 关于我
正在寻找 NLP / 大模型工程方向实习（北京 / 深圳 / 远程）。
擅长将研究想法落地为可复现、可评测、可部署的工程系统。
核心关注：中文拼写纠错（CSC）、由粗到细的误改控制（FPR）、模型可靠性。
A Chinese text correction system that prioritizes reducing over-correction (false edits) while maintaining strong correction quality.
Key Engineering Points:
- Fine-grained Error Typing: Distinguishing between phonetic (音) and shape-based (形) errors.
- Dual-Stage Pipeline: Candidate generation → Rescoring / Filtering.
- Type-Aware Calibration: Strict control over False Positive Rate (FPR).
中文说明： 一个重点降低“误改/过纠”（FPR）的中文拼写纠错系统。目前正在进行由粗到细的错误类型分类及两阶段流水线开发。计划在论文设计完成后开源。
| Project | Description | Tech / Topic |
|---|---|---|
| how-to-vibecoding | Vibecoding 系列教程：从环境搭建到多智能体协作的实战指南 | AI Coding MCP Tutorial |
| Volatility-Regime-Momentum | Quant Research: A-share momentum strategies under different volatility regimes. | Quant Finance A-Share |
| Codex-KBChat | Productivity: macOS menubar knowledge base + chat for local Markdown vaults. | Electron RAG Local-First |
| Predicting-medals | Data Science: GBRT-based Olympic medal count prediction (Paper + Code). | GBRT Data Mining |
| GAI-in-social-science | Survey: Trend-based review and resources for GAI in social science. | Survey Social Science |
I am available for freelance development and MVP prototyping. 寻求兼职开发、外包接单及产品孵化合作。
- MVP / Prototyping: Rapid demos for AI/NLP ideas (Backend + Simple UI).
- Data Engineering: Complex data cleaning, format conversion, and deduplication (SQL/Python).
- Automation: Workflow automation scripts and efficiency tools.
Connect with me via pingtianhechuan@gmail.com