Han Xiao Claim this profile

Telegram bridge for Claude Code

Give a query, get a dataroom. Pi + self-hosted Qwen3.6 research harness on a single L4.

Native macOS semantic search over your local files - text, images, audio, video in one vector space, on-device on Apple silicon.

Turn any document or a whole zip into an interactive knowledge graph, using a self-hosted Qwen3.6-35B-A3B-MTP on a single NVIDIA L4

Pure MLX implementations of UMAP, t-SNE, PaCMAP, TriMap, DREAMS, CNE, MMAE, and NNDescent for Apple Silicon. Metal GPU for computation and video rendering.

Deploy Qwen3.6-35B-A3B (Q4_K_XL) + MTP speculative decoding on a single NVIDIA L4 24GB — GCP g2-standard-8 — via the official llama.cpp Docker image. Decode-optimized to ~91–99 tok/s (min ~91 chat, max ~99 math), lossless (full GPU residency + ECC-off).

Some frequently used NLP blocks I implemented

DAANet: Dual Ask-Answer Network for Machine Reading Comprehension

Code for my blog post

Airgapped closed-corpus QA loop: a self-hosted Qwen3.6 agent explores a .zip dataroom under a token budget with local tools

UMAP in pure MLX for Apple Silicon. 30x faster than umap-learn.

Bridge incompatible embedding spaces with a single SVD. When your embedding provider deprecates a model, adapt instead of re-embedding.

A Tensorflow implementation of Yin Wenpeng's recent paper on TACL "Attentive Convolution"

Run Qwen3.5-35B-A3B with 1 million token context on a single NVIDIA L4

some useful one-liner

No description.

PaCMAP in pure MLX for Apple Silicon. Pure GPU, no scipy/numba.

MLX Local Serving (MLS) - Unified ASR, TTS, and Translation on Apple Silicon

IO-aware batched K-Means for Apple Silicon, ported from Flash-KMeans (Triton/CUDA) to pure MLX. Up to 94x faster than sklearn.

Deploy Qwen3.6-35B-A3B (Unsloth Dynamic Q4_K_XL) on a single GCP L4 24GB GPU with llama.cpp, Open WebUI, and nginx

Demo of building a flower image search using GNES Flow API

Poems retrieval demo built with GNES framework

Benchmarking GNES on its network latency

No description.

正确的读法是[huà xiá zi] 暂时懒得改了

Open-vocab multi-label image tagging with a single frozen model (jina-embeddings-v5-omni-nano-mlx). Zero training, no second model. Tokenizer-vocab labels + patch-level + CWR multi-crop test-time compute. mAP 0.710 on COCO-150.

An entire internet — search, pages & links — hallucinated on the fly by the local Qwen-AgentWorld-35B-A3B world model. Sequel to jina-ai/node-serp.

A shell-based wizard for GNES orchestration

Test-Time Compute for Dense Retrieval: Agentic Program Generation with Frozen Embedding Models

t-SNE in pure MLX for Apple Silicon. 12x faster than sklearn. FFT-accelerated (FIt-SNE) for O(n) scaling.

Real-ESRGAN on Apple Silicon via CoreML. 1.6x faster than MLX.

MindSpore is a new open source deep learning training/inference framework that could be used for mobile, edge and cloud scenarios.

CLA Assistant GitHub Action

A multilingual NLP annotation projection framework

System-prompt-driven knowledge-graph builder on a minimal Pi harness (Qwen3.6-35B-A3B-MTP on L4 + jina embeddings). Includes an A/B test: steps vs outcome prompting.

LLM inference in C/C++

A cool 3D hover effect for your awesome web project.

Open-source live chat software, an alternative to Intercom, Zendesk, Drift, Crisp etc. 🔥💬

Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more

App store search example, using Jina as backend and Streamlit as frontend

1.4B latent diffusion model fine tuning

Run GitHub Actions on self-hosted runner using Docker.

A Neural Net Training Interface on TensorFlow, with focus on speed + flexibility

TensorFlow code and pre-trained models for BERT

One interesting paper per day

Search engine for machine learning models and datasets

Lossless DFlash speculative decoding for MLX on Apple Silicon

Real-ESRGAN on Apple Silicon. Pure MLX, no PyTorch at runtime.

No description.

Universal memory layer for AI Agents

:sunglasses: Curated list of awesome lists

:closed_book: A minimal Markdown editor desktop app

Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞

FastAPI framework, high performance, easy to learn, fast to code, ready for production

DALL·E Mini - Generate images from a text prompt

A GitHub action to create a pull request for changes to your repository in the actions workspace.

No description.

GitHub Actions to generate TOC (Table of Contents)

Python collections that are backended by sqlite3 DB and are compatible with the built-in collections

No description.

Profile with dynamic realtime coding stats.

✨ Recognize all contributors, not just the ones who push code ✨

No description.

Image inpainting tool powered by SOTA AI Model. Remove any unwanted object, defect, people from your pictures or erase and replace(powered by stable diffusion) any thing on your pictures.

Textualize website

🦄 Record your terminal and generate animated gif images or share a web player

Slides & code for my talks

Qdrant - vector similarity search engine with extended filtering support

Official repository for the "Big Transfer (BiT): General Visual Representation Learning" paper.

Tool to help automate adding contributor acknowledgements according to the all-contributors specification ✨

GNES Hub ship AI/ML models as Docker containers and use Docker containers as plugins.

PyZMQ: Python bindings for zeromq

Quick tricks for troubleshooting TF

Log what you do on your computer. Simple (yet powerful), extensible, no third parties.

A complete and graceful API for Wechat. 微信个人号接口、微信机器人及命令行微信，三十行即可自定义个人号机器人。

No description.

Implement Coupled Generative Adversarial Networks in Tensorflow

Consulting service for PhD application in Germany

No description.

No description.

Blog

DSPy: The framework for programming—not prompting—foundation models

No description.

Fast and memory-efficient exact kmeans

Autonomous LLM-driven optimization of mlx-vis NNDescent on Apple Silicon

zvec-web is the official web frontend and documentation portal for Zvec

Minimal Nostr relay on Cloudflare Workers for OpenClaw agent-to-agent communication

A text-to-speech (TTS) and Speech-to-Speech (STS) library built on Apple's MLX framework, providing efficient speech synthesis on Apple Silicon.

Telegram Desktop messaging app

Telegram-iOS

Build Real-Time Knowledge Graphs for AI Agents

mini cli search engine for your docs, knowledge bases, meeting notes, whatever. Tracking current sota approaches while being all local

A lightweight, lightning-fast, in-process vector database

OpenViking is an open-source context database designed specifically for AI Agents. OpenViking unifies the management of context (memory, resources, and skills) that Agents need through a file system paradigm, enabling hierarchical context delivery and self-evolving.

No description.

MLX-VLM is a package for inference and fine-tuning of Vision Language Models (VLMs) on your Mac using MLX.

Plugins for Auto-GPT

🚴‍♀️ 阿里 - 很易用的中后台「表单 / 表格 / 图表」解决方案

No description.

Jina Executor: Convert image URI to blob

Jina V1 Official Documentation. For the latest one, please check out https://docs.jina.ai

A clean customizable documentation theme for Sphinx

Cookiecutter for building a new Jina executor

:crystal_ball: A lightweight comments widget built on GitHub issues

Bio

ZeroMQ Website

Django QuerySet like interface to query simple Python collections

Welcome to the one point community-driven encyclopedia for anything in technology.

Cookiecutter template for a Python package.

🌄Open Source AI Landscape: This interactive landscape similar to that of the CNCF Landscape (https://l.cncf.io/) that sorts through open source AI, ML and DL projects, and shows details including GitHub stars, funding or market cap, first and last commits, contributor count and many more.

Automatically close issues that don't follow the issue template.

Github Badge Maker

Serverless Framework – Build web, mobile and IoT applications with serverless architectures using AWS Lambda, Azure Functions, Google CloudFunctions & more! –

GitHub Action to send LGTM reaction

Convert your code to a dramatic banner

Asyncio (pep 3156) integration with ZeroMQ

Helm Chart

GNES docker images

📓 Lint commit messages

Open Source guidance from Zalando, Europe's largest online fashion platform

The source code that powers readthedocs.org

:symbols: A collection of GitHub issue and pull request templates

An Open Source Machine Learning Framework for Everyone

No description.

Under 1KB each! Super Tiny Icons are miniscule SVG versions of your favourite website and app logos

Vector icons, badges, flags etc. über optimised for web use.

Comment log for my blog

No description.

My solution to Machine Learning MNIST challenge 2017. [winner]

Swagger/OpenAPI First framework for Python on top of Flask with automatic endpoint validation & OAuth2 support

A framework for doing randomised UI testing of iOS apps

:trophy: Awesome, badges!

blog and notes

No description.

GloVe model for distributed word representation

Agency Theme for Jekyll

No description.

No description.

Device Registration Server for OJINS.COM News App

Website of Ojins.com app

Verkehrszeichen Guru

No description.

A swipeable cards interface. The swipe-left/swipe-right for yes/no input. As seen in apps like Jelly and Tinder.

nativeDroid2 is a jQueryMobile Template inspired by Material Design

a medium inspired jekyll theme

No description.