Building open-source products at the intersection of data engineering and AI.

把真实的数据工程、AI 工作流与研究问题，做成可运行、可审计、可复现的软件。

I build focused tools, self-hosted products, and research infrastructure around data engineering, metadata governance, reproducibility, and reviewable AI-assisted workflows.

A local-first research foundation for A-share fundamental-turnaround analysis. It builds a PIT-safe historical research corpus with raw-field provenance, canonical views, replay, evaluation, and explicit barriers against future information leakage.

Research infrastructure, not an investment-return or stock-picking product.

A self-hosted Feishu / Lark bookkeeping platform with Web and API entry points. Deterministic financial workflows, typed assistant responses, and query capabilities keep business actions reviewable; AI turns input into validated actions and does not directly operate the database.

A lightweight metadata management portal for data warehouses: discover and maintain data assets, field and table mappings, lineage, and governance-oriented metadata workflows. It is designed for real data-platform workflows and offline or self-hosted deployment.

A lightweight, framework-free, embeddable Web Component for interactive table and column lineage. It uses JSON, SVG, Shadow DOM, and zero runtime dependencies so lineage views can fit into existing applications.

Data Asset Portal and Audit are companion tools in the same data engineering toolchain: the public portal covers data assets, metadata, lineage, and governance workflows; Audit is the private companion for engineering audit. They are presented here as a product relationship, not as an already-integrated system.

• Data Asset Portal — public metadata and lineage workflows: data.overme.cn · source
• Audit — private, not open source: audit.overme.cn

Active contributor to a lightweight cross-platform database client. Recent upstream work spans SQL editor behavior, PostgreSQL correctness, schema diff, Kafka and data-transfer workflows, performance, frontend UX, and regression fixes.

My usual path is:

reproduce → isolate the root cause → make a scoped fix → add or strengthen regression coverage → send the change upstream

• Data infrastructure, metadata, and governance
• AI-native software engineering and agent workflows
• Point-in-time-safe quantitative research tooling
• Self-hosted products and lightweight automation
• Database tooling and upstream open-source contribution

• Inspectable by default — schemas, rules, assumptions, evidence, and generated artifacts should stay reviewable.
• Verification over vibes — reproduce → isolate → fix → regress; prefer evidence over guesses.
• Local-first where practical — keep sensitive metadata, credentials, workflow state, and usage data local or self-hosted when feasible.
• AI-native engineering — use Codex, Claude Code, Pi, and other coding agents as leverage while preserving acceptance criteria, tests, provenance, reviewability, and reproducibility.
• Small integration surface — prefer focused CLIs, APIs, and reusable components over an unnecessary all-in-one platform.
• Respect upstream — distinguish original work, forks, inspiration, and upstream contributions.

Python · TypeScript · Rust · Vue · React · FastAPI · PostgreSQL · SQLite · DuckDB · Parquet · Cloudflare Workers · D1 · Docker · LLM / agent workflows

• Website: www.overme.cn
• GitHub: github.com/0verme
• Data Platform: Asset Portal · Audit (Private, not open source)
• Lineage: lineage.overme.cn
• Ledger: ledger.overme.cn
• AI Usage: token.overme.cn