Xiaopei Tan

Technical Interests

• AI tooling & agent workflows: Tool calling, structured output validation, evals, and evidence-backed developer workflows.
• Full-stack systems: React and Next.js interfaces connected to backend APIs, auth, streaming routes, and deployment checks.
• C/C++ systems programming: Process control, POSIX APIs, async networking, desktop rendering, and cross-platform build pipelines.
• Async/runtime reliability: Non-blocking orchestration, real-time synchronization, deterministic checks, and production health paths.

Projects

• GitHub Issue Investigator. A small web app that lets an LLM read a GitHub issue and gather repo context without writing anything.
  Technologies: Next.js, TypeScript, React, AI SDK, OpenAI, NextAuth, Octokit, Zod, Docker, Render.
  • https://issue-investigator-dev.onrender.com/
  • The interesting bit is the tool layer: issue reads, comments, file reads, repo metadata, and code search.
  • I also added boring checks around schema shape and source links so the output is easier to trust.
• Unix Shell Implementation. A little Unix shell. Mostly an excuse to get closer to fork(), exec(), pipes, and signals.
  Technologies: C, Linux, POSIX APIs.
  • It runs foreground and background jobs with fork(), exec(), and wait().
  • It also handles pipelines, redirection, and basic signal behavior.
• Real-Time Multi-Room Chat Application. A class project chat app with a C++ backend, rooms, sockets, and a real database.
  Technologies: C++, React, PostgreSQL, Docker, Boost.Asio, WebSocket.
  • https://github.com/peixiaotan1/CS302_Final_Project
  • Boost.Asio handles the socket side; PostgreSQL stores the messages and account data.
  • The React frontend is plain, but it is enough to test the realtime room behavior.
• Cross-Platform Medical Image Viewer. A desktop viewer for poking at DICOM files and turning slices into a 3D surface.
  Technologies: C++, Qt6, OpenGL, DCMTK, CMake.
  • https://github.com/peixiaotan1/MedicalImageViewer
  • DCMTK handles the medical image files; Qt6 and OpenGL handle the viewer side.
  • The fun part is the marching-cubes path from image slices to a renderable mesh.

Paper

• Thinking While Driving: A Concurrent Framework for Real-Time, LLM-Based Adaptive Routing. A paper about letting simulated agents keep moving while the LLM is still thinking about routes.
  • https://arxiv.org/abs/2512.10610

Skills

Languages

TypeScript, JavaScript, Python, C, C++, C#, SQL

AI & Developer Tools

AI SDK, OpenAI-compatible APIs, Tool Calling, Prompt Design, Structured Output Validation, Evals, Octokit, Zod

Web/API

React, Next.js, Node.js, Flask, REST APIs, OAuth, NextAuth

Systems

Linux, POSIX APIs, Boost.Asio, Qt6, OpenGL, DCMTK, CMake

Infrastructure/Data

Docker, Render, GitHub Actions, PostgreSQL, MySQL, Microsoft Azure