Hi, I'm Ziniu. I build full-stack web apps and AI tools.

A Chrome extension: select text on any site and a floating panel lets you explain, summarize or rewrite it with OpenAI or Anthropic models, with the answer streaming in token by token — no copy-pasting into a separate chat window.

• Wrote an incremental Server-Sent Events parser that handles events split across network chunks; closing the panel aborts the in-flight request.
• Rendered the panel in a closed Shadow DOM so page CSS and scripts can't touch it, and model output as text, never HTML.
• Kept the API key in chrome.storage.local, used only by the service worker, with host permissions limited to the two model APIs.
• 26 unit tests and a GitHub Actions pipeline (type-check, test, build) on every push; zero runtime dependencies.

• TypeScript
• Chrome Extension (MV3)
• OpenAI & Anthropic APIs
• SSE streaming
• GitHub Actions

Browse restaurant menus, build a cart and check out. A Spring Boot REST backend with a React and Ant Design frontend, backed by PostgreSQL and running on AWS ECS Fargate.

• Built REST APIs for sign-up, menus, cart and checkout, layered into controllers, services and Spring Data JDBC repositories.
• Secured endpoints with Spring Security session-based authentication and BCrypt password hashing.
• Added Caffeine caching with per-customer eviction on cart updates; benchmarking 1,000 requests with the cache on vs. off halved median menu API latency (~1.8× sequential throughput, local database).
• Wrote 27 JUnit 5 / Mockito tests for the service layer, including cache tests that verify @Cacheable/@CacheEvict behavior.
• Containerized with a multi-stage Docker build; GitHub Actions CI runs backend and frontend tests and smoke-tests the running container on every push.
• Deployed to AWS ECS Fargate behind an HTTPS load balancer with a private RDS PostgreSQL database, with one-command deploy and teardown scripts.

• Java 21
• Spring Boot
• Spring Security
• PostgreSQL
• React
• AWS ECS & RDS
• Docker
• GitHub Actions
• JUnit & Mockito

Upload a document and every question gets two answers: one retrieved from the PDF, and one from a live web search run by a GPT-5 agent through an MCP server. Questions can be spoken, and answers read aloud.

• Built a RAG pipeline with LangChain: PDF loading, 500-character chunking, OpenAI embeddings and an in-memory vector store supplying context to GPT-5.
• Wrote an MCP server exposing a SerpAPI search_web tool over stdio, and a GPT-5 tool-calling agent that writes its own search queries, using the document title to resolve references like "this paper".
• Traced intermittent timeouts to SerpAPI stalling on about a third of uncached searches and added hedged requests; RAG and MCP run in parallel with isolated failures, cutting answer time from ~15s to 6–12s.
• Express backend for uploads and Q&A; React and Ant Design frontend with voice input and spoken answers via the Web Speech API.

• React
• Node.js
• Express
• LangChain
• GPT-5
• MCP
• SerpAPI
• Web Speech API

Post photos and videos, find posts with full-text search by keyword or author, or describe a picture and generate it with an OpenAI image model before posting. A Go backend on Google Cloud with Elasticsearch and Cloud Storage, and a React frontend on AWS Amplify.

• Built a Go REST API on Google App Engine, with Elasticsearch on a Compute Engine VM reachable only inside the VPC, media in Google Cloud Storage, and JWT-protected routes.
• Moved OpenAI image generation into a JWT-protected backend endpoint so the API key never reaches the browser, and migrated from the retired DALL·E 3 to the GPT Image API with the model made configurable.
• Fixed a signup race by creating users atomically in Elasticsearch (op_type=create, refresh=wait_for): an integration test fires 10 concurrent signups and gets exactly 1 account. Passwords are hashed with bcrypt.
• 42 Go tests (including integration tests against real Elasticsearch and a Cloud Storage emulator) and 26 Jest / React Testing Library tests, run by GitHub Actions on every push.

• Go
• React
• Elasticsearch
• Google App Engine
• Cloud Storage
• AWS Amplify
• OpenAI API
• JWT
• GitHub Actions