Golden Gate Bridge

AI Engineer / AI Researcher / Software Engineer

Luxin Zhang

I build reliable, production-grade AI systems and study how agents learn from experience, use interfaces, and operate safely in the real world.

Luxin Zhang

About

Building what’s next.

AI Engineer / AI Researcher / Software Engineer

My work spans lifelong agent memory, agent-facing interfaces, scientific discovery, digital health, secure systems, and full-stack product engineering. I turn research questions into evaluated systems and production-ready software.

Work

Ideas built into reality.

Two collections: competition-winning builds and the broader body of engineering work.

01

Competition work

Hackathon Winnings

6 award-winning projects

PatchPilot - Autonomous debugging engineer with Real PRsHonorable Mention

Honorable Mention

🏆 Honorable Mention - Build MCP Agents Hackathon

PatchPilot - Autonomous debugging engineer with Real PRs

PatchPilot is an autonomous debugging engineer that runs entirely inside an E2B sandbox. It reads your bug description or stack trace, analyzes your codebase, searches Exa for real-world fixes, generates a patch, validates it by running your tests, and finally opens a production-ready Pull Request through GitHub MCP, everything is all automatically!

Explore project ↗
02

Beyond competitions

Other Projects

30 projects · no hackathon duplicates

Call Me Tomorrow - Future-Self Voice Agent01

AI Projects

Call Me Tomorrow - Future-Self Voice Agent

Call Me Tomorrow lets Pipecat + Nemotron voice agents dial your future self, pulling from layered memory (ChromaDB + SQLite) to open with vivid scenes, weave in your fears, and deliver 30/90/365-day action plans, then auto-improves via Cekura eval loops.

View project ↗
SafeRun - visualizes cyber threats through animated AI guardians.03

AI Projects

SafeRun - visualizes cyber threats through animated AI guardians.

SafeRun transforms traditional cybersecurity dashboards into living, story-driven experiences. It visualizes real-time threats through animated AI guardians — from calm cats to defensive mecha dogs — blending game design with serious security analytics. By turning complex data into intuitive motion and narrative, SafeRun makes cyber defense both clear and captivating.

View project ↗
Vigil - Always Watching Your AI Stack04

AI Projects

Vigil - Always Watching Your AI Stack

Vigil combines SupplyGuard and AlignGuard modules to scan AI supply chains for .pth backdoors, credential harvesters, and C2 traffic while simultaneously catching prompt-injection attempts before they reach multi-agent pipelines, delivering a real-time behavioral security dashboard born from the March 2026 LiteLLM attack.

View project ↗
Warden - Zero-Trust Runtime Security Firewall for Agentic AI05

AI Projects

Warden - Zero-Trust Runtime Security Firewall for Agentic AI

Warden sits between agentic AI systems and the outside world, combining taint tracking, provenance ledgers, a neuro-symbolic supervisor, and a three-phase tool firewall to block prompt injection, tool hallucination, and unauthorized operations in real time while keeping memory writes clean and auditable.

View project ↗
VaultBrain - Collaborative VC Brain06

AI Projects

VaultBrain - Collaborative VC Brain

VaultBrain stitches GStack agents, GBrain persistence, and The Hog signals into a living VC brain so partners drop decks/emails, agents extract structured entities with provenance, scan live web intelligence, and keep a shared Brain Book timeline without humans typing CRM fields.

View project ↗
Memora - Ask your life, get the proof07

AI Projects

Memora - Ask your life, get the proof

Memora lets you ask any life-admin question by voice and get a concise answer plus the exact proof (PDFs, photos, notes). It highlights the supporting snippet, shows evidence thumbnails, and can set follow-ups (e.g., reminders before expiration). All data stays private by default.

View project ↗
SupportBrain - Autonomous Support Engineer08

AI Projects

SupportBrain - Autonomous Support Engineer

SupportBrain ingests customer mail, runs dual memory layers, fans out to Hyperspell + Nia for cross-source context, classifies CODE_BUG vs inquiry, opens GitHub issues with PatchPilot-ready context, and can even ship PRs—closing the full support loop instead of stopping at a reply.

View project ↗
Demoxa - AI-Powered Automated Demo Generator09

AI Projects

Demoxa - AI-Powered Automated Demo Generator

Demoxa parses your README with Gemini, plans a DemoScript, then drives your live app via a vision-based computer-use agent running on Playwright to record professional walkthrough videos with optional manual capture fallback and live monitoring.

View project ↗
CardSmart - Real-Time Credit Card Advisor10

AI Projects

CardSmart - Real-Time Credit Card Advisor

CardSmart pairs a FastAPI backend, React dashboard, and n8n orchestration to recommend the smartest credit card for each purchase, monitor weekly budget burn, and surface personalized rewards using Minimax LLM reasoning and promo-matching workflows.

View project ↗
Éclat - SeedDance AI Ad Studio11

AI Projects

Éclat - SeedDance AI Ad Studio

Éclat with SeedDance turns curated product stills plus a short brand brief into 12-shot art-directed storyboards and cinema-grade 9:16 ads, preserving exact packaging while auto-publishing optimized cuts to TikTok Shop, Instagram Reels, and YouTube Shorts in ~90 seconds.

View project ↗
ARIA - Autonomous Reasoning & Intelligence for Ads12

AI Projects

ARIA - Autonomous Reasoning & Intelligence for Ads

The Automatic multi-agent Advertisement Systems involving ARIA (Autonomous Revenue Intelligence Agent) is an AI-powered advertising optimization system that autonomously manages multi-platform ad campaigns through continuous experimentation, learning, and adaptation. The system uses LangGraph orchestration to coordinate specialized agents that observe signals, generate strategies, create content, optimize audiences, allocate budgets, execute campaigns, and evaluate results.

View project ↗
CHRONOS - Multi-Agent Collaboration Platform15

AI Projects

CHRONOS - Multi-Agent Collaboration Platform

CHRONOS is an open-source multi-agent collaboration platform that enables AI agents to discover each other by skills, share structured context safely, propose actions, debate decisions, and resolve conflicts through enforceable AI Court rulings

View project ↗
Random Forest with the Spam Data24

Neural Network & Statistics

Random Forest with the Spam Data

Split the data randomly into training (75%) and testing (25%), first build the best random forest to predict Spam e-mails using the training data, then use the out-of-bag (OOB) data to measure its performance, and then use this random forest model to predict whether each e-mail in the testing data is Spam or not.

View project ↗
Logistic Regression25

Neural Network & Statistics

Logistic Regression

Used all the other 8 variables to predict the binary dependent variable 'diabetes' telling us whether the subject is diabetic or not (factor with 2 levels: neg and pos). You will split the data into 80% training and 20% testing.

View project ↗
Battleship Game27

Game

Battleship Game

Built a full-stack web application called Battleship Game, which allows two players and one server. Players can move, attack, and perform sonar scan.

View project ↗
Traveling App28

iOS App

Traveling App

Built a traveling app that tested on IOS The advantage of this application is that it can give you all the recommendations for Restaurants, Attractions, and Hotels separately within one search each. (Search based on Google Map API)

View project ↗
Open the complete project index ↗

Open source

Build in the open

Hypothesis ATLAS

AI-powered evidence mapping system for biotech research that discovers, organizes, and visualizes signals from papers, datasets, and scientific updates.

View repository ↗

Research

Accepted poster · 3rd Workshop on Lifelong Agents (AIAS 2026)

From Episodes to Procedures: Scope-Aware Memory Promotion for Lifelong Enterprise Agents

Luxin Zhang · November 7, 2026

ScopeAware turns successful agent experiences into shared procedures only when the evidence is reliable and the scope is explicit. Across enterprise support and scientific-facility evaluations, it improves resolution accuracy while sharply reducing harmful memory promotion.

View paper →

Accepted poster · VISA 2026 at IEEE/ACM MICRO 2026

What Makes a Virtual ISA Agent-Programmable? A Controlled Study of Agent-Facing Documentation for StableHLO

Luxin Zhang · November 2026

A controlled study of how documentation format and delivery affect AI agents programming StableHLO. Syntax alone produces large gains, while matched prose outperforms JSON contracts and upfront documentation consistently beats on-demand retrieval.

View paper →

Academic service

Reviewing and judging

Reviewer and program committee service across artificial intelligence, agent evaluation, computational linguistics, digital forensics, and public health.

Reviewer

Third Workshop on Synthetic & Adversarial Forensics (SAFE 2027)

SAFE Workshop

November 2026

Official website ↗

Program Committee Member (Reviewer)

AAAI-27 Undergraduate Consortium

Association for the Advancement of Artificial Intelligence

October 8–14, 2026

Official website ↗

Reviewer

2027 ASPPH Annual Meeting for Academic Public Health

Association of Schools and Programs of Public Health

October 2026

Official website ↗

Reviewer

EMNLP 2026 Industry Track

Association for Computational Linguistics

July 22–24, 2026

Official website ↗

Contact · Let’s build what’s next

Have an idea worth
building together?

I’m always open to thoughtful conversations about AI systems, research, engineering, and ambitious products.