Language Breakdown
Lines of code distribution across 53 owned repositories
M-Shaped Developer
M-shapedMulti-specialist across TypeScript, Jupyter Notebook, Python
Collaboration Network
Global Impact visualization
Repos
91
PRs
0
Growth
+18%
Top Collaborators
No collaborator data yet.
Coding Streak
Contribution activity over the past year
WeiHaoran
@Ucas-HaoranWei
Vincent Koc
@vincentkoc
Vignesh
@vignesh07
Radek Sienkiewicz
@velvet-shark
Shadow
@thewilloftheshadow
Top Repositories
Welcome to the Pi Agent Improvement Suite! This repository provides enterprise-grade extensions, integrations, and core patches for the earendil-works/pi coding agent. It drastically enhances Pi's security, operational reliability, and ecosystem connectivity.
AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI
The Dynamic Real-Time Fraud Detection System is an end-to-end, production-ready machine learning platform designed to detect and block fraudulent financial transactions in under 50 milliseconds.
A production-grade, distributed microservices platform for Enterprise Semantic Search and Retrieval-Augmented Generation (RAG). Capable of ingesting millions of documents, indexing them via Hybrid Search (BM25 + Dense Vectors + custom SPLADE v2), and serving sub-100ms LLM-powered answer generation.
LiteLlama is a high-performance, from-scratch Large Language Model inference engine tailored for the LLaMA-3 model family.
Sparse-Attention provides custom CUDA/Triton kernels that replace the standard O ( N 2 ) dense attention mechanism with a structured block-sparse pattern. By combining heterogeneous attention constraints (local sliding windows, global stride landmarks, and prefix context), it enables Large Language Models to scale context lengths up to 1M tokens
A production-grade Serverless Pastebin infrastructure deployed on AWS using Terraform. Designed to be fully scalable, highly observable, and strictly secure while leveraging the AWS Free Tier.
Development repository for the Triton language and compiler
The Triton-Based Fused Operator Suite targets the most critical bottlenecks in Large Language Model (LLM) inference pipelines. Written purely in OpenAI Triton, these kernels bypass PyTorch's ATen overhead by aggressively fusing operations, maintaining data in highly performant SRAM, and dramatically reducing Global Memory (GMEM) round-trips.
Open Source Impact
Contributions to external projects
