Repository constellation

Plus 6 smaller infrastructure experiments.

PRODUCTION AI · VANCOUVER

Yichen Wu

I build production AI across computer vision, generative AI, and the full-stack systems around them.

Tap a bright repo. Dim nodes trace earlier builds.

VIEW WORK ↓RESUME ↗ (opens in new tab)
github · 5 flagship + 8 dim buildsVANCOUVER PT

SELECTED WORK

Systems that prove
their work.

A growing set of open-source systems showing how I build reliable AI products across testing, research, agent safety, developer tools, and production infrastructure. Each project links directly to its code and documentation.

PRODUCTION SCOPE

reusable AI workflow nodes shipped
20+
persistent IDs for traffic analytics
TRACKING
persistent jobs with progress and failover
ASYNC
graph-aware Pydantic workflow checks
VALIDATION
01 · OPEN SOURCECASE STUDY →

Argus

An MCP QA system that turns coding agents into exploratory test engineers for web and native macOS apps—probing workflows, state, and UX for bugs scripted tests were never written to find.

SYSTEM FLOW

Test strategy → exploratory testing → clean-state reproduction → CI replay

DESIGN

  • Semantic interactions · browser · network · state · visual evidence
  • Clean-reload receipts · model-free replay · JUnit · SARIF

WHY IT MATTERS

Finds behavioral and product bugs beyond predefined test cases, then produces reproducible evidence engineering teams can trust.

MCP · PYPI · PLAYWRIGHT · macOS

02 · OPEN SOURCECASE STUDY →

Sibyl

A web research system for AI agents that performs keyless multi-source retrieval, ranks passage-level evidence, and can produce citation-backed reports through a configured model.

SYSTEM FLOW

Question → multi-source search → evidence ranking + deduplication → cited report

DESIGN

  • Keyless retrieval · passage-level provenance
  • Syndicated-copy clustering · explicit insufficiency states

WHY IT MATTERS

Goes beyond one-pass model browsing by grounding research reports in ranked, traceable evidence and distinguishing repeated copies from independent support.

RESEARCH · MCP · CLI · PYTHON

03 · OPEN SOURCECASE STUDY →

Loop Agent

An autonomous software delivery runtime that takes a product brief, plans and builds the implementation end to end, tests and repairs its work, then hands back reviewable software with proof that the agreed requirements pass.

SYSTEM FLOW

Product brief → acceptance contract → plan + build → verify + repair → delivered implementation

DESIGN

  • ReAct execution · shell · browser · files · Git change sets
  • Independent verifier · required artifacts · Apply / Discard / Undo

WHY IT MATTERS

Turns a request into working, reviewable software instead of stopping at generated code or a model's claim that it is done.

AUTONOMOUS DELIVERY · AGENT RUNTIME · GIT · VERIFICATION

04 · OPEN SOURCEGITHUB ↗ (opens in new tab)

Scalable Starter

A reusable Next.js and FastAPI foundation connecting a web app, API, background workers, Postgres, Redis, and Kubernetes.

SYSTEM FLOW

Next.js → FastAPI → Postgres

FastAPI → Redis queue → worker

DESIGN

  • At-least-once jobs · ack · retry · dead-letter recovery
  • Typed contracts · migrations · observability · validated overlays

WHY IT MATTERS

Gives teams a production-shaped starting point instead of rebuilding service boundaries, queue reliability, and deployment plumbing.

NEXT.JS · FASTAPI · POSTGRES · KUBERNETES

05 · OPEN SOURCECASE STUDY →

Duet

A local desktop collaboration loop where Codex implements, Claude Code independently reviews, and machine checks decide whether work is ready.

SYSTEM FLOW

Isolated worktree → read-only review → machine checks → guarded apply

DESIGN

  • Single writer · finite review rounds
  • Repository fingerprints · exact-state undo · transcript-free receipts

WHY IT MATTERS

Adds independent review without letting either model overwrite the developer's main workspace or bypass failed checks.

ELECTRON · MCP · GIT · AGENT SAFETY

OTHER BUILDS

Clinic Web Platform

CLIENT DELIVERY · PHP · APACHE · SEO

A production website and lightweight content system for a Vancouver medical-aesthetics clinic, with responsive treatment pages, bilingual navigation, custom PHP publishing tools, site search, and automated canonical URLs, sitemaps, and redirects.

Website Knowledge Chat

PRIVATE PROTOTYPE · NEXT.JS · RAG · PGVECTOR

A multi-tenant platform that turns client websites and uploaded documents into searchable knowledge bases, retrieves relevant passages with pgvector, and stores conversation history plus contact details visitors choose to share.

Content Network

PRIVATE SYSTEM · CONTENT PIPELINES · AUTOMATED QA

A regional publishing system that ingests public data, generates location-specific pages and search indexes, and runs automated duplicate, image, link, and HTML checks before deployment.

Wander

PRIVATE PROTOTYPE · NEXT.JS · FASTAPI · MAPS

An evidence-backed AI travel planner that verifies places against live map data and combines LLM guidance with travel-time geometry to build realistic, editable multi-day itineraries.

TRAJECTORY

From Wuxi to
production AI.

Three countries, two engineering degrees, and a path from mobile and IoT into full-stack delivery, computer vision, and generative AI.

  1. — 2018WUXI, CN

    Wuxi

    Grew up in a canal city an hour from Shanghai. School, exams, and the first itch to build things.

  2. 2018 — 2022URBANA-CHAMPAIGN, US

    Urbana-Champaign

    Crossed the Pacific at eighteen for UIUC. After freshman year, spent summer 2019 in Shenzhen building an Android and Bluetooth smart-lock prototype.

    B.S. COMPUTER ENGINEERING · DEAN'S LIST

    SUMMER 2019 INTERNSHIP · ANDROID · BLUETOOTH · IOT

  3. 2022 — 2023EVANSTON, US

    Evanston

    Master's at Northwestern, straight through.

    M.S. COMPUTER ENGINEERING · NORTHWESTERN UNIVERSITY

  4. 2024 — 2025CHICAGO, US

    Chicago

    Deployed SageArk's Next.js site on AWS and built a SageMaker job-recommendation system from feature engineering through demo integration.

    FULL-STACK DEVELOPER · NEXT.JS · AWS · SAGEMAKER

  5. 2025 — NOWVANCOUVER, CA

    Vancouver

    Building production video analytics and full-stack generative AI workflows at DaoAI Robotics, from graph-aware validation and multi-object tracking to persistent job state, progress updates, and retry/failover.

    SOFTWARE ENGINEER · VIDEO ANALYTICS · GENERATIVE AI

3 COUNTRIES · 5 CITIESAWS SOLUTIONS ARCHITECT ASSOCIATE · CLOUD PRACTITIONER
FULL RESUME ↗ (opens in new tab)

OPEN TO THE RIGHT TEAM

Let's build AI that
holds up.

Based in Greater Vancouver, I'm exploring AI and full-stack engineering roles and am open to relocation for the right team. If you're building production systems where reliability, thoughtful system design, and measurable results matter, I'd like to hear from you.

EMAIL MERESUME ↗ (opens in new tab)
yichenwujob@gmail.comlinkedin (opens in new tab)resume.pdf (opens in new tab)github.com/chriswu727 (opens in new tab)