GITHUB PAGES PORTFOLIO

Pinned Projects

Open-source work by @alanthssss — AI research platforms, developer tooling, and collaborative resources.

Featured Projects

6 pinned repositories

sweetgold

Python
Repository Live Site

A reproducible multi-agent AI research lab built around a deterministic bee-colony simulator. Combines a matched-seed benchmark, an interactive Strategy Arena, and an honest robustness audit pipeline — supporting behavior cloning, PPO, and centralized training with decentralized execution (CTDE) while keeping training, validation, and test seeds strictly isolated.

Objective
Provide a self-contained, hardware-portable environment for comparing multi-agent RL policies under reproducible, seed-isolated conditions.
Key Engineering
  • Current stage M16 phase 1: all ML commands support explicit CPU / Apple MPS / NVIDIA CUDA device selection — no silent fallback; run manifests record the hardware backend for honest cross-machine audit.
  • M6 BC+PPO and M7 CTDE smoke workflows validated on Apple M1 Pro GPU.
  • Matched-seed arena enforces identical environmental conditions across policy comparisons.
Stack
  • Python
  • PyTorch
  • PPO
  • CTDE
  • TypeScript
  • JavaScript
  • HTML
License
Apache 2.0
  • behavior-cloning
  • ctde
  • multi-agent-reinforcement-learning
  • ppo
  • pytorch
  • reproducible-research
  • simulation

collaboration

Docs
Repository Live Guide

An open-source bilingual playbook for software teams and open-source communities. Provides a one-page Collaboration Protocol, practical guides for discussions, meetings, feedback, and retrospectives, plus copy-ready Markdown templates — so every session ends with a decision, an owner, and a date.

Objective
Turn circling discussions into clear decisions and accountable owners by giving teams a minimal shared protocol and ready-to-use templates.
Key Engineering
  • Published as a live guide and downloadable PDF at alanthssss.github.io/collaboration/.
  • Full bilingual (English / Simplified Chinese) coverage across all templates and guides.
  • Situation-indexed template table lets teams jump directly to the relevant template (discussion, meeting, decision, retrospective).
License
CC BY 4.0
  • bilingual
  • collaboration
  • decision-making
  • playbook
  • templates
  • retrospective
  • teamwork

mcata · Merge Boost

TypeScript
Repository

A themeable merge roguelike puzzle game — and a full AI benchmark framework — built with React, Vite, and TypeScript. Players merge tiles on a 4×4 grid across escalating stages, collecting Boosts and navigating Hazards, while the underlying numeric engine serves as a reproducible benchmark target for AI agents.

Key Engineering
  • Pluggable theme system drives tile visuals independently of the numeric merge engine, enabling cosmetic changes without breaking benchmark compatibility.
  • Run-level Rule modifiers (chosen before each run) vary corner bonuses, spawn frequencies, output scaling, and step budgets.
  • Six Style archetypes with in-run upgrade paths enable deep build diversity.
  • Hazard stages (Entropy Tax, Collapse Field) stress-test both player and AI strategies.
Stack
  • TypeScript
  • React
  • Vite
  • CSS
License
GPL-3.0
  • benchmark
  • game
  • reactjs
  • roguelike-game

pw2048

Python
Repository

An automation-first benchmark and experimentation platform that uses the 2048 game as a compact, deterministic environment for comparing search and reinforcement learning algorithms. Designed as a small engineering system covering the full model lifecycle — training, evaluation, registry, quality gates, deployment, observability, and rollback.

Objective
Demonstrate that AI experiments can be compared, reproduced, released, and operated safely — connecting Env → Train → Eval → Registry → Quality Gate → Deploy → Observe → Rollback in a single coherent system.
Key Engineering
  • Models carry immutable manifests; candidates must pass offline quality checks before promotion.
  • Stable/Canary workloads expose metrics and traces, run in isolated Kubernetes environments, and support automated rollback.
  • Multi-interface developer experience: CLI, TUI, and Web frontends.
  • Dockerized, with Shell-scripted automation covering the full benchmark workflow.
Stack
  • Python
  • Playwright
  • Docker
  • Kubernetes
  • Shell
  • HTML
  • JavaScript
License
GPL-3.0
  • algorithms
  • benchmark
  • playwright
  • reinforcement-learning
  • performance

petal

Go
Repository

A setup-friendly, Ansible-inspired lightweight distributed infrastructure operations tool. Designed for ease of adoption with a single binary distribution and minimal configuration overhead.

Key Engineering
  • Ships as a single binary, removing runtime dependencies for distributed ops tasks.
  • Dockerfile and Makefile included for reproducible builds and container deployment.
Stack
  • Go
  • Shell
  • Docker
  • Makefile
  • ansible-like
  • distributed
  • golang
  • infra
  • lightweight
  • ops

pulse

Go
Repository

A lightweight, cross-platform CLI tool for monitoring network service connectivity. Provides infrastructure and middleware health-check capabilities from the command line.

Key Engineering
  • Cross-platform: runs on Linux, macOS, and Windows via Batchfile scripting.
  • Go Template-driven output for flexible, scriptable reporting.
Stack
  • Go
  • Shell
  • Go Template
  • infrastructure
  • middleware
  • network
  • cli

Notes

Preserved reference material

Python Notes

Language basics, execution modes, and virtual environment notes.

Docker Notes

Containers, images, Compose, and networking walkthroughs.