Skip to content
View wuwangzhang1216's full-sized avatar
🎯
Focusing
🎯
Focusing
  • San Francisco

Organizations

@Lightning-Goods

Block or report wuwangzhang1216

Report abuse

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

Report abuse
wuwangzhang1216/README.md

Hey, I'm Steve Wu 👋

AI Engineer & LLM Researcher

Building AI-native tools that actually ship. Obsessed with making large language models more useful, more controllable, and more open.


🚀 What I'm Working On

  • ��� OpenYak — Open-source local-first AI desktop app supporting 100+ models
  • 🔬 Abliterix — Fully automated LLM abliteration framework. LoRA + Optuna TPE optimization, 135+ model configs, 9 peer-reviewed techniques (NeurIPS/ACL/ICLR). 0–1.5% refusal rate with 0.01 KL divergence.
  • 🧠 LLM alignment research — abliteration, representation engineering & steering vectors
  • 🔍 Vector search engines & RAG pipelines at scale
  • 🤖 Multi-agent orchestration — built from scratch, no heavy frameworks
  • 📝 AI-native writing & knowledge systems
  • 🔬 Publishing fine-tuned & abliterated models on HuggingFace

🛠 Tech Stack

Languages
Python C++ Rust Go CUDA TypeScript

AI/ML & Deep Learning
PyTorch Transformers HuggingFace PEFT DeepSpeed vLLM

Research & Techniques
RepEng Abliteration RLHF Quantization FlashAttention KnowledgeDistillation MoE

LLM APIs & Inference
OpenAI Anthropic Ollama

Vector Databases & Retrieval
FAISS pgvector Milvus

System Design & Distributed Systems
Microservices DistributedSystems EventDriven APIDesign Kafka RabbitMQ LoadBalancing Caching Sharding HA Scalability Observability

Full Stack
FastAPI Next.js React PostgreSQL Redis

Infra
Docker Kubernetes AWS Linux


🎓 Background

  • 🎓 Honours BSc in Computer Science & Mathematics, University of Toronto (3.95/4.0)
  • 💼 10+ years in Databases, LLMs & AI Agent Systems

HuggingFace

Pinned Loading

  1. openyak/openyak openyak/openyak Public

    Open-source local-first AI agent for desktop work. No account, no telemetry: use local models with Ollama/Rapid-MLX or bring your own provider key.

    Python 703 59

  2. abliterix abliterix Public

    Automated alignment adjustment for LLMs — direct steering, LoRA, and MoE expert-granular abliteration, optimized via multi-objective Optuna TPE.

    Python 221 35

  3. claude-code-source-all-in-one claude-code-source-all-in-one Public

    Always up-to-date open-source mirror of Claude Code (currently v2.1.152). Run from source with Claude subscription/API, ChatGPT subscription (GPT-5.5 / GPT-5.4), OpenAI-compatible providers, or loc…

    TypeScript 90 93

  4. nexu-io/html-anything nexu-io/html-anything Public

    ✨ The agentic HTML editor — your local AI agent writes the HTML, you ship it. 🚀 75 Skills × 9 Surfaces (magazine · deck · poster · XHS / tweet · prototype · data report · Hyperframes) 🛡️ Sandboxed …

    HTML 8.6k 844

  5. ChinaTextbookStudyFree ChinaTextbookStudyFree Public

    给所有中国小学生打造的免费学习平台,从一年级到六年级所有的课本知识点全汇总。

    Swift 140 41

  6. DirectorSKILL DirectorSKILL Public

    Claude Code skill for AI filmmaking — shot lists, keyframe & video prompts, with 10 director-style overlays (Spielberg, Kubrick, Wong Kar-wai, Nolan...).

    71 17