Skip to content
View AluminumShark's full-sized avatar

Highlights

  • Pro

Block or report AluminumShark

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
AluminumShark/README.md

Hi, I'm Levi 👋

AI Engineer | Agent Systems · Security · Cloud Solutions

I build AI systems that go beyond demos — from multi-agent applications and RAG pipelines to cloud deployments and tools for testing agent security.

My background spans enterprise AI delivery, causal inference research, and open-source engineering. I enjoy turning ambiguous requirements into practical systems, understanding why something works (or doesn't), and making complex technology easier for others to use.

Outside work, I maintain Python packages and teach hands-on AI workshops at NTU AI Club.

LinkedIn Website PyPI

What I'm Focused On

Agent security & evaluation — Exploring how to verify that AI agents respect tool permissions, and how to turn authorization checks into repeatable engineering tests.

Enterprise AI solutions — Designing and delivering agent workflows, retrieval systems, and cloud-based AI applications while working across technical and customer-facing teams.

Open source & technical education — Building tools, publishing reproducible examples, and teaching practical agent engineering to the community.

Featured Projects

lockwalk — Authorization Testing for AI Agents

PyPI Total Downloads

An alpha-stage Python toolkit for testing the authorization setup around AI agent tools. It identifies untested permissions, compares expected decisions against an existing policy engine, checks whether denials are enforced, and can surface unexpected agent-to-tool routes from traces. Designed to make these checks visible in CI — not to replace IAM or act as an authorization gateway.

pip install lockwalk

LangGraph Skills Workshop — Teaching Agent Engineering

A hands-on NTU AI Club workshop I co-developed and teach, covering LangGraph workflows, tool calling, Agent Skills, and tool-permission safety. Includes runnable Google Colab notebooks, slides, practical exercises, and an English edition.

Synthetic Difference-in-Differences — Open-Source Causal Inference

PyPI Total Downloads

A Python package implementing the SDID estimator for causal inference and policy evaluation, built out of my quantitative research interests and published on PyPI.

pip install sdid

Carbon Emissions Trading Experiment — Real-Time Market Simulation

A multi-player experimental economics platform built with oTree, featuring a live carbon-permit market, WebSocket-based price updates, and an order-matching engine. Connects economic research design with concurrent, event-driven software to study market behavior and emissions-trading mechanisms.

More Things I've Built

FinFlash — Multi-agent financial-news analysis with structured outputs and retrieval.

Social Debate AI — LLM debate system combining RAG with experimental GNN/RL modules and ablation studies.

Build My Own Redis — Redis-like C++ key-value server exploring TCP sockets, non-blocking I/O, and event loops.

Technologies I Work With

Area Tools & methods
AI & agents LangGraph, RAG, MCP, Agent Skills, PyTorch, LLM APIs
Cloud & backend GCP (Vertex AI, Cloud Run, Cloud SQL), FastAPI, PostgreSQL / pgvector, Docker, IBM watsonx
Security & observability Agent authorization testing, IAM concepts, OpenTelemetry, Langfuse, LangSmith
Programming & research Python, C++, SQL, causal inference, optimization, experimental design

Background & Community

I studied political science at National Taiwan University, worked in research at Academia Sinica, and moved into AI engineering. I now enjoy work that combines hands-on implementation, technical problem-solving, and clear communication with the people who use the systems we build.

I also share practical notes and tutorials on tensorOwl.

Pinned Loading

  1. Synthetic_Difference_in_Difference Synthetic_Difference_in_Difference Public

    A Python implementation of Synthetic Difference-in-Differences (SDID) for causal inference and policy evaluation.

    Python 3

  2. lockwalk lockwalk Public

    Authorization test coverage for AI agent tools: find untested tool permissions, policy mismatches, and unenforced denials.

    Python 2

  3. Carbon-Emissions-Trading-Experiment Carbon-Emissions-Trading-Experiment Public

    An oTree platform for studying how carbon reduction policies—such as trading, taxes, and controls—affect producer behavior.

    HTML 2 2

  4. langgraph-skills-workshop langgraph-skills-workshop Public

    NTU AI Club 工作坊:從 LangGraph 到 Skills,結構化與非結構化 Agent 的雙拼指南(Colab notebooks+簡報)

    HTML 13

  5. ibm-cpd-onprem ibm-cpd-onprem Public

    A portable CLI toolkit and Claude Code skill for managing IBM's on-prem AI platforms.

    Python

  6. Social-Debate-AI Social-Debate-AI Public

    A deep learning-based multi-agent social debate system that integrates RAG, GNN, and RL technologies for intelligent debate simulation.

    Python 6