I build small AI workflow tools that turn repeated business work into reusable Skills, scripts, reports, and human-review checklists.
中文：我关注 AI 工具在业务流程、文档、表格和运营复盘中的落地，重点是把高频重复任务拆成标准输入、执行步骤、输出模板、检查清单和人工审核边界。
Workflow demo based on ecommerce operations. It shows how order, inventory, and support-ticket data can be turned into risk reports, action queues, and review boundaries.
- Pages: https://onyx679.github.io/ecommerce-ops-ai-workflow-kit/
- Repository: https://github.com/onyx679/ecommerce-ops-ai-workflow-kit
- Release: https://github.com/onyx679/ecommerce-ops-ai-workflow-kit/releases/tag/v0.1.4
What it demonstrates:
- SKU risk scoring from simulated order, inventory, and support-ticket inputs.
- Markdown weekly risk reports and action queues.
- Clear separation between AI-generated drafts and human-reviewed business decisions.
- Transferable workflow thinking from ecommerce operations to business-process AI enablement.
Public portfolio project for automotive value engineering productivity workflows.
- Pages: https://onyx679.github.io/automotive-ve-ai-skills-kit/
- Repository: https://github.com/onyx679/automotive-ve-ai-skills-kit
- Reviewer case study: https://github.com/onyx679/automotive-ve-ai-skills-kit/blob/main/docs/case-study.md
- Release: https://github.com/onyx679/automotive-ve-ai-skills-kit/releases/tag/v0.4.5
What it demonstrates:
- AI Skill design for scenario mining, VAVE opportunity drafts, SOP writing, adoption feedback, evidence audits, and community Skill adaptation.
- Python CLI tools for CSV inputs, Markdown outputs, scoring, readiness checks, and claim evidence matrices.
- GitHub Actions tests, bilingual README, GitHub Pages, release discipline, and explicit human-review boundaries.
Small, merged documentation contributions:
- genai-io/san: documented
make cias the local PR gate in the contributing guide. PR: genai-io/san#238 - harvard-cns/orla: added the
orlactlsource build command to the quickstart. PR: harvard-cns/orla#54
Open PRs are not listed here as evidence until they are merged.
- Decompose messy workflows into repeatable inputs, steps, outputs, and review gates.
- Prefer small, inspectable tools over vague AI promises.
- Write documentation and examples so reviewers can verify the work quickly.
- Keep claims tied to public evidence.
The featured portfolio projects use simulated or non-confidential data. They do not use internal company material, real supplier quotations, real vehicle BOMs, proprietary vehicle-program data, real customer records, or private merchant exports.