Pierre-Antoine Bannier

bark.cpp. Suno AI's Bark model in C/C++ for fast text-to-speech generation

sam3.cpp. Fast state-of-the-art image and video segmentation in portable C/C++

encodec.cpp. Fast neural codec compression and generation for audio waveforms

WSIStreamer. A tile server for WSI stored in S3-compatible object storage.

biogpt.cpp. Port of Microsoft's BioGPT in C/C++ using ggml

nanograd. A torch-like lightweight deep learning framework around Numpy arrays

PathCollab. A self-hosted collaborative digital pathology viewer with real-time cursor presence and overlays

nanokv. A distributed key value store in ~1,000 lines of code.

sparseglm. Fast and modular solver for sparse generalized linear models

cs_notes_tips_tricks. A bunch of notes about CS, ML, networking that I usually learn on the job.

NarrateMate.ai. TypeScript

automatic_hp_selection_for_meg. Automatic hyperparameter selection for Lasso-like models solving the M/EEG source localization problem

claude-telegram-bot. A Telegram bot to send and receive prompts to Claude Code.

PathView. PathView is a high-performance whole-slide image viewer for digital pathology. PathView combines GPU-accelerated tiled rendering with polygon overlays for cell segmentation visualization and AI agent integration via the Model Context Protocol (MCP).

MakeItPunchier. Make your tweets more punchy with Mixtral and GPT

fgfr3mut. Official Implementation of the "AI allows pre-screening of FGFR3 mutational status using routine histology slides of muscle-invasive bladder cancer" paper

rust-ggml. Rust bindings for C tensor library ggml

inria_bilevel_optim. HTML

learning_assembly. you find C hard? learn assembly then C will feel easy peasy

pabannier.

benchmark_multi_task_lasso. Benchopt benchmark for Multi-Task Lasso

fast-melanoma-detection. A malignant melanoma classifier with deep learning deployed in a Streamlit web app

Vector-field-visualizer. A simple tool to visualize a vector field written with Processing

sinkhorn-swift. Benchmarking mask interpolation methods for real-time on-device video object segmentation on Apple Silicon. 11 methods compared (optical flow, optimal transport, baselines) across 30K+ evaluations — with Metal compute shaders, Apple Vision flow, and Sinkhorn OT.