Trending archive for Oct 13, 2026

openai/math

OpenAI's model-generated math proofs, with Lean formalizations in progress — A collection of 719 mathematical manuscripts in 372 families produced by an internal OpenAI model tackling open research problems, released as model-development evaluations saturated on existing benchmarks. About 42% of top-line results have Lean formal proofs so far, with reasoning summaries and a comparator tool included for verification.

rust-lang/rust

The core repo behind Rust's compiler and standard library — This is the main source for the Rust compiler, standard library, and docs. It bundles tooling like Cargo, rustfmt, Clippy, and rust-analyzer, aimed at developers wanting memory and thread safety without sacrificing performance.

Empowering everyone to build reliable and efficient software.

openai/codex

OpenAI's terminal-native coding agent, now with IDE and app options — Codex CLI runs locally in your terminal as a coding agent from OpenAI, installable via curl, npm, or Homebrew. The same Codex experience extends to VS Code/Cursor/Windsurf via an IDE plugin, a desktop app, or a cloud-based version at chatgpt.com/codex.

Lightweight coding agent that runs in your terminal

h0m3us3r/w1700k-sdr

Custom firmware turns a Wi-Fi 7 router into a 320 Msps SDR — Replaces MediaTek's firmware on a Gemtek W1700K router with reverse-engineered code, exposing its MT7996 radio as a wideband SDR streaming 14-bit IQ at up to 320 Msps over 10 GbE. A Linux host renders a live spectrum/waterfall in-browser, aimed at RF hobbyists and researchers wanting cheap wideband capture across 2.4/5/6 GHz without vendor code.

franzenzenhofer/big-arrow-on-the-screen

A CLI that lets AI agents literally point at your screen — bigarrow is a macOS CLI and Claude Code/Codex skill that draws a click-through arrow, box, or text over any window so an AI agent can show you exactly what to click. It needs no macOS permissions, runs as a single Swift binary with no daemon or telemetry, and disappears on its own.

Let your AI agents paint big arrows, boxes and text on your Mac screen. One CLI, click-through, gone by itself. Skill for Claude Code and Codex. MIT.

mattn/yomivec

Phonetic Japanese word search powered by SQLite's vec1 — A Go CLI that finds Japanese words by sound using SQLite's vec1 vector extension, with no language model involved. It hashes phonetic 'sound keys' into vectors for cosine-distance search, accepting English, romaji, kana, or kanji queries. Useful for anyone building IME-like lookup tools or exploring lightweight vector search with SQLite.

Find Japanese words by sound with SQLite vec1

oboroge0/hayamimi

CPU-only multilingual live transcription that beats Whisper on Japanese CER — hayamimi routes each utterance to a specialist quantized ONNX model per language via sherpa-onnx instead of one general Whisper model, hitting 3.8% CER on Japanese broadcast audio vs Whisper-large-v3-turbo's 13.8%, at 10-50x realtime on a 6-core CPU, under 2GB RAM, with no GPU or cloud API.

早耳 - Real-time multilingual speech-to-text on CPU only. Live subtitles, browser dashboard, speaker labels, translation. No GPU, no cloud.

apache/seatunnel

SeaTunnel extends data integration to multimodal and LLM workloads — Apache SeaTunnel syncs huge data volumes across 160+ connectors, now covering video, images, and unstructured text alongside structured data for embeddings and LLM pipelines. It runs on Zeta, Flink, or Spark engines, supporting batch, streaming, and CDC with a distributed snapshot algorithm for consistency. Useful for teams needing flexible, multi-engine data pipelines.

SeaTunnel is a multimodal, high-performance, distributed, massive data integration tool.

todesking/slopsampler

Cataloging the telltale phrases of AI-generated Japanese text — This project systematically generates Japanese text across multiple themes and language models, then analyzes the output to catalog word- and sentence-level patterns typical of 'AI slop' — like overused 'not A but B' constructions. Useful for writers, editors, or researchers studying how AI-generated Japanese differs from natural prose.

スロップすぎるよ展

nekorobiaws-maker/jev-graphrag-poc

GraphRAG without embeddings, driven by score thresholds — A proof-of-concept GraphRAG system that skips vector search and embeddings entirely, using Jev (TypeSafe AI) scores and thresholds to decide entity registration, graph traversal entry points, and when to stop hopping. Claude Haiku 4.5 is used only for final answer generation from the gathered evidence chunks. Built on AWS with DynamoDB, Lambda, and CDK.

エンベディングを使わない判断特化型GraphRAGのPoC（判定はJev、回答はClaude Haiku 4.5）