Xiaopei Tan

Technical Interests

• Production web platforms: Next.js and TypeScript apps with server-side validation, bilingual content, admin tooling, and the boring parts that keep them running.
• Backend reliability & data: PostgreSQL schema design, transactions, transactional outboxes, concurrent job claims, retries, and recovery paths.
• Cloud infrastructure: AWS Lambda containers, Terraform-provisioned environments, keyless GitHub Actions OIDC deploys, and object storage with presigned uploads.
• C/C++ systems programming: Process control, POSIX APIs, concurrency and async I/O, desktop rendering, and cross-platform build pipelines.

Projects

• Tan Handy Memphis. A bilingual (EN/中文) lead platform I built and shipped on my own for a home-renovation contractor. It is in production: marketing pages, a server-validated submission pipeline, and an admin console.
  Technologies: Next.js, TypeScript, PostgreSQL, Drizzle, Cloudflare R2.
  • https://tanhandymemphis.com
  • Notifications go out through a transactional outbox, so they commit with the lead and drain asynchronously; workers claim rows with an atomic SKIP LOCKED grab, plus capped retries and stale-claim recovery.
  • Job photos are processed in the browser first (HEIC transcoding, EXIF orientation baked into the pixels, metadata stripped so home GPS never leaves the device), then uploaded straight to a private R2 bucket with a presigned PUT.
• GitHub Issue Investigator. An issue-investigation tool built on read-only Octokit queries, where every claim links back to the source it came from.
  Technologies: TypeScript, Next.js, AWS Lambda, Terraform, Zod, Octokit, Docker.
  • https://w4p7hynbmrtpu7ruag466nfyei0qqliv.lambda-url.ap-northeast-1.on.aws/
  • Streamed LLM output is validated against a Zod report schema, so malformed chunks get dropped before render and permission or rate-limit failures come back as typed results.
  • It runs on Lambda containers behind a response-streaming Function URL to keep SSE intact, provisioned as 12 Terraform resources and shipped through keyless GitHub Actions OIDC.
• Unix Shell Implementation. A little POSIX shell in C. Mostly an excuse to get closer to fork(), exec(), pipes, and signals.
  Technologies: C, POSIX APIs, Linux.
  • Process control with fork/exec/wait, plus background job control.
  • Multi-stage pipelines with explicit file-descriptor wiring, I/O redirection, and signal handling.
• Cross-Platform Medical Image Viewer. A desktop viewer for poking at DICOM files and turning slices into a 3D surface. Ships for Windows, macOS, and Linux.
  Technologies: C++, Qt6, OpenGL, DCMTK, CMake.
  • https://github.com/peixiaotan1/MedicalImageViewer
  • A DCMTK parsing pipeline handles the medical imaging formats, including the compressed ones.
  • The fun part is the marching-cubes path from image slices to a mesh rendered through OpenGL.

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

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

Backend & Data

Flask, Node.js, Next.js, React, PostgreSQL, MySQL, Redis, REST APIs, Schema Design, Transactions

Cloud & Infrastructure

AWS Lambda, Docker, Terraform, GitHub Actions, Cloudflare R2, Vercel

Systems & Tools

Linux/POSIX, Concurrency, Async I/O, Git, CMake, Qt6, OpenGL