Weici (Will) Zhao

Software Engineer · Full-stack development, AI agents and RAG

About

I’m a software engineer and an M.S. student in AI for Data Science at Penn. I build full-stack products and AI agents: review and audit automation at Apple, a table-analysis agent at an applied-AI startup, and on the side NotchPal, a macOS app that lives in the MacBook notch.

What I care about most is whether the system holds up. Most of my projects ship with an evaluation set, a fallback path and a number I can point to.

Experience

Apple · Sales, via a third-party agency

AI Engineer Intern · Beijing, China · Mar 2026 – Sep 2026

  • Built a human-in-the-loop review agent that automated about 95% of submission reviews, with no false approvals across 90+ validation cases.
  • Built frontend workflows and backend integrations for a retail experimentation platform: A/B-test assignment, submission intake, monitoring and automated reminders, with dealer and store data queried in SQL.
  • Developed JD.com competitive-audit automation with XCUITest, OCR and vision LLMs, covering 12 accessory categories and 25+ product-search scenarios, with fallback paths for UI and search failures. Added a supporting web dashboard and extended the audit to U.S. Amazon with a scraping workflow.
  • Built a local-first spreadsheet agent that splits and merges semi-structured Excel workbooks using schema inference and validated execution plans, with an evaluation harness, plan caching and an API for agent-to-agent orchestration.

DICA · Applied AI startup

AI Engineer Intern · Shanghai, China · Mar 2025 – Jul 2025

  • Built a tool-using ReAct agent for tabular data analysis with plot generation and natural-language interpretation, +7 points accuracy over the o4-mini baseline on DABstep’s Easy split.
  • Found stale-memory leakage across context boundaries in role-playing LLM agents and built context masking to limit reuse of information outside the intended context.
  • Built a role-playing LLM evaluation pipeline (persona extraction, question templates, cross-model voting) and curated a 100-persona Pantheon-derived dataset to compare self-consistency and contradiction rates.

Supremind

Software Engineer Intern, CAD Automation · Hangzhou, China · May 2024 – Jul 2024

  • Developed a C# automation tool on Siemens NX Open APIs with a parameter-entry interface, turning customer modelling workflows into repeatable sequences that generate complete models in one click.

Projects

NotchPal · A pixel pet and widgets in the MacBook notch

macOS app · 25+ widgets · notchpal.xyz

  • A native macOS app that lives in the MacBook notch. Hover over it and a panel unfolds: a pixel-art pet wanders through a hand-drawn scene in the middle, with your widgets on either side.
  • 25+ widgets on a grid, from 1×1 icons to 5×3 panels (now playing, calendar, weather, translation, Pomodoro, AI usage meters), with layouts saved as presets that sync across Macs.
  • Drop zones on the notch to AirDrop, shelve, trash, zip or convert files, and a slim system HUD that replaces the stock volume, brightness and keyboard-backlight overlays.
  • Social rooms: visit friends and your pets share one scene, walk together, or sit around a campfire and chat. Mini-games run right inside the notch.
  • Built with Swift and SwiftUI (IOKit, EventKit, CoreLocation, on-device Translation), Supabase for accounts, realtime rooms and sync, Stripe and Sparkle. A one-command release pipeline runs the tests, signs and notarizes the build and updates the website. Developed entirely with Command Line Tools, without Xcode.

WeChat2All · Extensible agent control plane

Open source · GitHub

  • An open-source, local-first agent integration platform that connects mobile messaging to GUI agents, CLIs, backend APIs and MCP tools.
  • A plugin protocol and SDK for isolated, hot-installable integrations: tools and workflows can be added or replaced without modifying or redeploying the core.

Legal Manus · RAG-powered legal Q&A

Hit@3 68% → 98% · Website

  • A deployed RAG legal Q&A app grounded in official documents, with FastAPI streaming APIs and a React frontend for interactive research.
  • Indexed 600K+ legal passages with text-embedding-3-small and added a BGE cross-encoder reranker, raising Hit@3 from 68% to 98% on a 200-question evaluation set.

Education

University of Pennsylvania

M.S. in AI for Data Science · GPA 3.78 / 4.0 · Philadelphia, PA · Expected May 2027

University of Virginia

Dual bachelor’s degrees in Computer Science and Applied Statistics · GPA 3.91 / 4.0 · Charlottesville, VA · Aug 2021 – Dec 2024

Skills

Languages
Python, TypeScript, SQL, Java, C++, C#, Swift
Full-stack & data
React, Node.js, FastAPI, Spring Boot, REST APIs, PostgreSQL, MySQL, Redis, Supabase
AI & ML
PyTorch, Transformers, LangChain, RAG, Tool calling, Embeddings, Vector search, Reranking, LLM evaluation, Vision LLMs, OCR, MCP, scikit-learn, pandas, NumPy
AI coding tools
Codex, Claude Code, Cursor, GitHub Copilot
Cloud & DevOps
Docker, AWS EC2, Alibaba Cloud OSS/RDS, Git, CI/CD, Linux