NLP

Native MLX reimplementation of the Mamba-3 architecture from state-spaces/mamba, adapted for Apple Silicon (MLX). The implementation mirrors the block-selection logic, core Mamba-3 recurrence concepts, inference-cache semantics, and config structure, while adding MLX-native generation, benchmarking, serialization, and LoRA fine-tuning utilities.

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

LLM inference in C/C++

LLM과 Tensor Network(MPS)를 활용한 멀티 에이전트 기반 신약 후보 탐색 플랫폼

GoldWorm 🐛✨ A 302-neuron cognitive engine in Zero-Trust Rust, inspired by C. elegans. No backpropagation—just Dual-Stream Architecture and Hebbian Echo Resonance for real-time temporal learning. Sparse 302-D manifolds, synaptic decay, and an ANSI mission dashboard. Processes 200K+ words/sec on one core.

Natural-language end-to-end GUI testing for Windows. Write test cases in plain English; a perception layer sees the screen and a reasoning model acts them out.

From-scratch language models in Swift: n-gram, neural autograd, a MLX GPT, and RAG.

Turn natural language into anime prompts using a local LLM. No cloud, no API keys.

Local-first video timelines for AI agents: transcribe, read on-screen text, and query videos over MCP.

Security Assessment: Sarvam-105B Identity Fragility, Prompt Robustness Failures, and Information Disclosure (8 vulnerabilities, CVSS 4.0-7.5)

LLM inference in C/C++

Tunable compression scheme for dense embedding indexes

Dashboard de rescate que organiza un canal de Telegram de emergencias con IA. PWA + API pública. https://angelesdelaautopista.com

A transparent, reproducible benchmark of frontier LLMs against real Congressional Research Service bill summaries.

AI-powered n8n workflow that collects customer feedback, identifies feature requests, prioritizes them with Gemini AI, logs results to Google Sheets, and handles errors automatically.

Local semantic search engine in Rust: BERT embeddings, binary index, SIMD search, polling monitor, TUI

A creative directive framework for AI: imagination over fact, story over safety, depth over polish. 15 rules for fiction, worldbuilding, brainstorming, and exploration — built to amplify originality, narrative coherence, and tonal courage. Designed to be used as a system prompt. Give your AI the permission to speculate, contradict, and craft boldly

Extensión Chrome para fact-checking en tiempo real de debates políticos chilenos

JAX implementation of "Knowledge Distillation of Black-Box Large Language Models" (Chen et al., 2024)