I build full-stack AI products and developer tools with Java, Vue, Python, RAG, and MCP.
Beyond the model response itself, I care about the data behind it, how results are checked, which actions require human approval, and how a deployed system can be tested, debugged, and rolled back.
Safely adds Claude Code configuration and skills to an existing project with backups, incremental merges, and rollback.
Python Claude Code CLI Backup Rollback
A CI-first evaluation tool for AI applications, with versioned cases, deterministic metrics, reviewed baselines, quality gates, and reproducible reports.
Python Evaluation CI JSONL MIT
A local-first research workspace that keeps claims traceable to sources, captured passages, and report history.
FastAPI MCP Playwright SQLite Web Research
A self-hosted Spring Cloud and Vue 3 commerce platform with a shopping assistant grounded in the current product catalog.
Java Spring Cloud Vue 3 RAG Docker
- Connecting domain models, interfaces, APIs, data, and deployment into a complete product.
- Grounding RAG and MCP systems in real data rather than stopping at a chat demo.
- Reducing change risk through tests, logs, backups, and rollback.
- Keeping important model-driven actions behind explicit human approval.
Most portfolio repositories are now public. Public visibility does not imply an open-source license or production readiness; each case study still separates implementation evidence, maturity, demo availability, and remaining work.