Skip to content
View eauchs's full-sized avatar
🎯
Focusing
🎯
Focusing
3 followers · 3 following

Achievements

Achievement: Pull Shark

Achievements

Achievement: Pull Shark

Block or report eauchs

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

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

Report abuse
eauchs/README.md

Théophile Lafargue

I build local AI infrastructure on Apple Silicon.
Open-source contributor to llama.cpp and StepFun. Patent holder.
Student-Entrepreneur at Paris-Saclay (Pépite PEIPS).

Projects

Project What it does
Phantom Personal AI OS — behavioral transformer that observes, learns, and acts. MLX on-device.
VoxTape Local voice-to-text for macOS — MLX Whisper on Metal GPU. Open-source SuperWhisper alternative.
llama.cpp #20075 Fix speculative decoding on hybrid SSM/MoE models. +45% inference on M3 Max.
StepFun Cookbook #14 Local deployment of Step-3.5-flash on Apple Silicon. Merged.
benchmark-422-qec 11 LLMs vs. one quantum error correction problem. 0/11 correct.
patent-low-bandwidth-ai Hybrid RAG pipeline for AI over 2G/SMS — backend for patent FR2511116.
gui-agent Two-layer macOS GUI agent — VLM perception + LLM strategy.
enigma-shell Control a full Linux VM with natural language via local LLMs.

Stack

Inference — llama.cpp · MLX · Metal · ROCm · GGUF
AI/ML — Whisper · RAG (ChromaDB) · LoRA · Transformers
Languages — Python · C++ · TypeScript
Infra — Docker · Tailscale · Bare-metal homelab (7 devices)

LinkedIn · Substack · HuggingFace

Pinned Loading

  1. enigma-shell enigma-shell Public

    An experimental web shell to control a full Linux OS (v86) with natural language via local LLMs.

    JavaScript 3

  2. speech-to-speech-pipeline speech-to-speech-pipeline Public

    A real-time, interruptible (barge-in) conversational AI pipeline (STT-LLM-TTS) running locally. Optimized for Apple Silicon (MLX).

    Python 2

  3. gui-agent gui-agent Public

    A two-layer GUI agent for macOS. A VLM (Vision Language Model) handles perception, while a separate LLM (Qwen) manages high-level strategy and decision-making. Built with Python, OpenAI API, and Py…

    Python 1

  4. patent-low-bandwidth-ai patent-low-bandwidth-ai Public

    Backend for my 'Stateful AI over Low-Bandwidth Networks' patent (FR2511116). A hybrid RAG pipeline (SmolDocling, ChromaDB, Reranker) with SMS support, separating local VLM (perception) from the mai…

    Python 1